US 20020079425 A1 Abstract An imaging system for measuring the field variance of distorted light waves collects a set of short exposure “distorted” images of an object, and applies a field variant data processing methodology, resulting in an image estimate which approaches the diffraction limited resolution of the underlying physical imaging system as if the distorting mechanism were not present. By explicitly quantifying and compensating for the field variance of the distorting media, arbitrarily wide fields can be imaged, well beyond the prior art limits imposed by isoplanatism. One embodiment comprehensively eliminates blurring effects introduced when imaging through the earth's atmosphere.
Claims(3) 1. In a method of imaging a subject from a sensor array, at least a portion of the earth's atmosphere being interposed between the subject and the sensor array and thereby tending to introduce distortion in imaging of the subject, an improvement characterized by characterizing an aspect of said distortion over a field of view exceeding 10 arc seconds, and compensating therefor. 2. The method of 3. The method of Description [0001] The present invention relates to the imaging of objects which have been distorted or blurred, and more particularly relates to a method and apparatus for creating clear wide field images after first characterizing and then compensating for the distortion. The distortion is characterized by being both spatially and time variant. [0002] The limitations on imaging system performance imposed by a turbulent media, most simply described as ‘blurring,’ are well known, particularly in applications using medium to large aperture telescopes in the open atmosphere. These limitations have not only led to a variety of system solutions that will be discussed as prior art, but have played a major role in the decision to launch space based telescopes and have led to serious postulations of lunar based observatories. [0003] For a large aperture telescope—generally greater than a 10 centimeter diameter for the visible light region—which is otherwise constructed to a precision commonly referred to as “near diffraction limited,” the overall ability to resolve objects obscured by a turbulent atmosphere is limited by the turbulence rather than by the instrument For the visual band of light once more, it is quite typical for a 1 meter aperture telescope to have ten times worse resolving power due to the turbulence, while a 10 meter aperture telescope can be 100 times or more worse than its innate “diffraction limit.” The exact numbers for any given telescope on any given night are a function of many variables, but this general level of degradation is widely recognized. As importantly, this atmospheric blurring directly leads to a loss in effective sensitivity of these large aperture imaging systems, which either renders dim objects just too dim to be seen or forces greatly extended exposure times, ultimately limiting the number of objects that can be imaged during a given length of usage time. [0004] The prior art for addressing this problem and trying to alleviate it can be generally categorized into the following well known areas: 1) Telescope Placement; 2) Adaptive Optics Systems; and 3) Speckle Interferometric Systems. It would not be unfair to say that the system disclosed herein is best summarized as a fundamental expansion to the third category, though this is only in a most general sense. [0005] Regarding the first category, that of simply finding locations of minimal turbulence, the placement of telescopes at high elevations has been known and practiced since Isaac Newton's era. This typically adds some expense, but more critically, it is well known that this only goes a small way toward reaching diffraction limited performance, and moreover, imaging performance is still quite subject to the varying atmospheric conditions of a given site. The space age has brought about the “obvious” solution of launching telescopes above the atmosphere; though at considerable expense and complexity. [0006] The second category of adaptive optics has been well known for decades and has seen significant physical realizations over the last two decades. A brief technical summary of such a system is that after a telescope primary mirror has collected the light waves emanating from a given object, it splits the light wave into two “beams.” One beam goes into an instrument known as a wavefront sensor and the other beam enters an ordinary imaging system or camera. The wavefront sensor derives information about the phase distortion (caused by the atmosphere) of the incoming light wave and in less than hundredths or even thousandths of a second, sends control signals to mirror actuators which advance and retard primary beam mirror surfaces in cancelling opposition to the incoming phase distortion There are two critical problems with these systems, however. First, they are expensive to build and expensive to maintain. Second, they can only increase the resolving power within an extremely small angle about the nominal central (paraxial) ray of the telescope, typically on the order of 2 to 10 arc seconds in the visible band. This “runnel vision” is technically referred to as the “isoplanatic patch.” It arises due to the fact that the phase distortion of a three dimensional media such as the atmosphere changes rapidly as a function of field angle. [0007] The third category of speckle interferometry was begun and developed roughly contemporaneously with adaptive optics systems, i.e. over the last two decades. The guiding principle of these systems is to take a large number of short exposure images of an object, thereby “freezing” the phase distortion (in time) of a given single exposure, and ultimately deriving the fourier magnitude and the fourier phase of the object being imaged through mathematical operations on the accumulated set of exposures. The state of the art of these techniques now includes the addition of a wavefront sensor to an overall system, which provides additional information to the computational operations, allowing for a smaller number of short exposures to be gathered to achieve a given level of precision and also simplifying the overall process. Though these speckle systems can be less expensive than adaptive optical systems, they too are limited by “tunnel vision” of the isoplanatic patch. Moreover, since they do not account for the field variance of the distortion, they suffer from higher noise and error levels, which in turn affect their effective sensitivity, accuracy, and throughput (number of images needed for a given level of accuracy). [0008] The expense and limited performance of the collective prior art has accordingly limited its application and hampered its broader acceptance. A novel aspect of the system disclosed herein is that it will quantify and compensate for the field variant complex phase distortion across an arbitrarily wide field of view, at a cost much less than a comparable adaptive optics system, and with a throughput and accuracy much better than speckle interferometric systems. In so doing, it can also provide advantages over these prior art systems even in applications where the object of interest is contained within the so-called isoplanatic patch; most notably in cost per image at a given accuracy (error and signal to noise ratio). [0009] Just as adaptive optics systems have recently employed “artificial beacons” to assist in the imaging of very dim objects, so too can the systems described herein utilize various forms of this concept as described herein. Artificial beacons can be employed when the brightness of an object under study is insufficient or inappropriate to provide photons to a wavefront sensor. The beacons are generally laser beams directed along a close line of sight to the object, generating backscatter photons which will undergo largely similar phase distortions as the photons from the object under study, and thus they can be used to deduce the phase distortions applicable to the object photons. [0010] The preferred embodiment of the system disclosed herein is a large telescopic imaging system used in the open atmosphere. This is chosen as an important and tangible example of obtaining clear images through a general field variant distorting media. It should be evident that other physical instruments, subject to distorting mechanisms other than an atmosphere, can also produce sets of digital images and corresponding sets of field variant distortion estimates, thereby being capable of utilizing the claimed methods of processing these sets of digital data into a single clear image. [0011]FIG. 1 pictorially summarizes some of the principal components of a system according to a preferred embodiment of the present invention. [0012]FIG. 2 is a schematic overview of the physical components of the preferred embodiment. [0013]FIG. 3 is a graphic representation of some of the mathematical foundations which are used to generate specific implementations of the processing methodology. [0014]FIG. 4 is a detailed schematic of camera system embodiment using a simple wide field Hartmann camera embodiment. [0015]FIG. 4A shows a typical array of sub-apertures placed at the prime focal plane of the FIG. 4 embodiment. [0016]FIG. 4B is a schematic representation of spot images as they might appear on the Hartmann detector in the FIG. 4 embodiment, with a further blow-up to a depiction of the centroids of those spots during typical turbulent conditions. [0017]FIG. 5 is a schematic functional drawing of a wide field wavefront sensing system according to one embodiment of the invention. [0018]FIG. 6 is a schematic representation of a field angle multiplexing beam splitter. [0019]FIG. 6A is a schematic view of the mirrors in the mirror sub-unit of FIG. 6, as seen by the incoming light beam. [0020]FIG. 6B is a front view of the two checkered mirrors of the FIG. 6 embodiment. [0021]FIG. 6C is a schematic drawing of the field stop sub-apertures which can be used in a 2X wide field Hartmann embodiment. [0022]FIG. 7 is a schematic view of a wide field Hartmann camera according to one embodiment of the invention, called the “full” wide field Hartmann camera. [0023]FIG. 8 is a schematic drawing of the optical train of one embodiment of the invention, along with a list of example specifications which are used throughout the disclosure in order to further explain detailed design issues. [0024]FIG. 8A is a schematic drawing of the 16 round pupil images as they would appear at the lenslet plane in front of the wavefront detector, with the lateral displacements of the beamspliners overlaid. [0025]FIG. 9 is a hardware and computer system block diagram of an illustrative embodiment. [0026]FIG. 10 is an overview of the processing methodology used in an illustrative embodiment. [0027]FIG. 11 is a representation of a generic weighted average operation. [0028]FIG. 11A is a logical step overview of the role that the weighted averaging operation can play in mitigating the traditional difficulties involved with image estimation based on inversions. [0029]FIG. 12 is a schematic diagram of the centers of field angles at which the wavefront is measured, overlaid on a grid of solution pixels. [0030]FIG. 12A are examples of spectral sensitivity curves of the light beams which enter the primary camera (on the left) and the wavefront camera (on the right). [0031]FIG. 13 is a schematic representation of the illustrative embodiment of the overdetermined basis functions used in the processing methodology. [0032]FIG. 13A is a representation of how these basis functions are arranged inside computer memory in an illustrative embodiment. [0033]FIG. 14 is a diagram showing a processing methodology according to an illustrative embodiment of the invention. [0034]FIG. 15A is an assistance schematic, clarifying some of the process steps of FIG. 14. [0035]FIG. 15B performs a similar clarification on another programming item. [0036]FIG. 16 displays some of the general characteristics of “the border problem” referenced herein. [0037]FIG. 17 is a processing step overview of illustrative methods which determine image plane distortions based on Hartmann spot data. [0038]FIG. 18 is a schematic which takes a closer look at the image data derived from the full wide field Hartmann camera, showing two levels of zooming in, culminating in a general look at nominal spot brightness overlap and crosstalk. [0039]FIG. 18A looks at “in operation,” i.e. turbulent conditions, spot crosstalk and overlap. [0040]FIG. 19 is a detailed description ofthe processing step of the global wavefront sub-routine, which is a sub-routine in the main program of the illustrative embodiment. [0041]FIG. 19A is a description of the processing steps involved in a local centroiding sub-routine [0042]FIG. 19B is a description ofa sub-routine which estimates the point spread function from a complex pupil function. [0043]FIG. 20 is an intuitive graph which illustrates the relation between “obvious” applications of prior art methods (1A) to six novel methods of this invention (1, 2, 3, 4, 1B, & 1C). [0044]FIG. 21 is dctailed description of the processing steps of the point spread function method of creating “rev_array,” an array used in the main program. [0045]FIG. 21A is a description of the local wavefront sub-routine, corresponding to the point spread function method. [0046]FIG. 22 is a graphical depiction of the relationship of novel processing methods of this invention to the “obvious” application of prior art processing methods. [0047]FIG. 23 is detailed description of the processing steps of the data point window function method of creating “rev_array,” an array used in the main program. [0048]FIG. 23A is a description of the local wavefront sub-routine, corresponding to the data point window function method. [0049]FIG. 24A is a schematic diagram assisting in the explanation of the data point window method of FIG. 23, depicting in particular the mirroring of indices between psf array and the window array, in the lower right two drawings. [0050]FIG. 24B graphically provides further insight into the operation of the processing steps of FIG. 23, shown as they might affect the H matrix in equation 1 of the text. [0051]FIG. 25 is a schematic with extended explanatory text in the figure, outlining the rationales of refining the methods of detetrmining rev_array for use in the main program reversion process. [0052]FIG. 26A is a schematic which highlights optical spatial filtering effects to which the wide field wavefront camera described herein will generally be prone to. The lower figures detail the effective fourier optics operation perfortned by the field sampling aperture. [0053]FIG. 26B schematically outlines the framework in which the effects of the spatial filtering are mitigated by the processing methodology of FIG. 27. [0054]FIG. 27 is a preferred embodiment of the processing methodology which can mitigate the errors that would otherwise have been introduced by the spatial filtering of the sampling apertures of FIGS. 26A and 26B. [0055]FIG. 28 is a general overview of other refinements to the data processing. [0056] FIGS. [0057]FIG. 30 is a simple example of how a 2 by 2 lensiet array might image a grid in typical atmospheric distortion. [0058]FIG. 31 is a process flow chart for largely removing the effects of scintillation in a distorted image. [0059] FIGS. [0060] FIGS. [0061] FIGS. [0062] FIGS. [0063] FIGS. [0064]FIG. 37 is the process flow of removing the higher order distortion effects of a single detector system. [0065] The disclosure begins with a general nescription of ithe aggregate of parts of the system and then will discuss each part in detail. FIG. 1 contains a schematic summary of the central aspects of the preferred embodiment. The physical entities are enumerated and the process flow will be described in the text, In FIG. 1, an object [0066]FIG. 2 has an overview of the physical systems and sub-systems of the preferred embodiment It is perhaps most convenient to split the physical components of the system into those elements whicn attach directly to the telescope and those which are part of an operating station. This disclosure concisely refers to this distinction as the camera system [0067] As mentioned in the discussion surrounding FIG. 1. the physical system in FIG. 2 will collect “E speckle exposures” in order to obtain one clear image [0068] In order to better understand what signal to noise segregating basis functions are and why they are used, we can briefly skip ahead to FIG. 13 which has a simplified schematic depicted of the ovcrdetermined basis functions used in the preferred embodiment, with one in particular blown up. This figure shows that the final image estimate is broken down into a generally dependent (overdetermined) set of basis functions which maintain generally “knowable” signal to noise characteristics. Singular value decomposition techniques perform a more precise method of segregating the image into an independent set of basis functions, unique to each atmospherically distorted wavefront, whereas here, we merely want to approximate this segregation in a simpler, computationally efficient manner which is not unique to each exposure, hence the use of overdetermination. The purpose of this step is so that each individual primary image exposure can separate out its unique signal to noise components into a common framework so that it can add its own properly weighted estimate to an averaged final image estimate. The basis functions chosen for the preferred embodiment are, effectively, the global fourier components AS CONTRIBUTED BY local data regions. [Computational efficiency will show that we don't need to use absolute global fourier coefficients, however, hence only the 256 by 256 local fourier components]. In this manner, a given exposure with a given local region which has a global “null” frequency will add no weight to a final image estimate, while another local region with a good signal t [0069]FIG. 13 shows the primary detector plane [0070] Continuing with FIG. 13, each exposure is in turn broken down into these data segments, the segments are stepped through, the frequency domain representation of the local window function—as determined by the processing of the wide field wavefront sensor data—is multiplied by the frequency domain representation of the data in the segment, and the result is accumulated into three image estimate data planes: the real numerator, the imaginary numerator, and the magnitude denominator. [FIG. 1 treats the complex numerator as one entity.] FIGS. 11 and 11A concisely describe the rationale behind these data planes, showing them to be complex components of a typical weighted averaging process. The more detailed discussion surrounding FIGS. 11 and 11A will show that the adverse theoretical and practical effects of null components, which are quite well known in the art of image restoration, is entirely removed at this point due to the cancellation of potential divisions by zero by an equal multiplication of that zero in the service of the weighted average. In brute force deconvolution terms, the acquired signal component would require a division by the magnitude of the transforming component (which could easily be a zero, see step [0071] Finishing up with FIG. 13 and FIG. 1, then, all exposures are processed in this manner, an overall image estimate is represented as the accumulated real and imaginary numerators and magnitude denominators, each segment is then stepped through where the division of the complex numerators by the single denominator takes place to give a local UV plane estimate, then this estimate is either directly added to the accumulating final image UV plane estimate, or alternatively, is fourier transformed into the spatial domain and added into the accumulating final object estimate in that domain. After all segments are cycled through, the result is a clear image [0072] Now that the general description of an illustrative system is completed, it will be more readily apparent how and why the subcomponents function as they do. These items will now be described in detail. [0073] The Camera System [0074]FIG. 4 depicts a more detailed optical layout of the camera system [0075] The primary camera light beam [0076] The preferred embodiment of the wide field wavefront camera is what can be called a “basic wide field Hartmann camera” [0077] Referring back to FIG. 4. the shuttering device [0078] Now that a simple but slightly limited form of a wide field wavefront sensing system has been described in the context of FIGS. 4, 4A, and [0079] A fundamental variant to the simple wide field Hartmann design of FIG. 4 is what can be referred to as the beam splitting wide field Hartmann camera. It is entirely probable that of the two specific beam splitting designs about to be described, the second or “full field” design will be the most commonly utilized in physical systems even though it is the most complicated and expensive, its performance gain will usually be worth it. Therefore, of these, the full field design is presently preferred, while the “simple” design presented in FIG. 4 and the “2X” design about to be presented will be viable alternatives to the full field design. (A third implementation is described at the end of this specification.) [The 2X design in particular may some day get very close to the full field ideal, especially when yet further refinements are made to “descrambling” the information gathered by the wavefront detector [0080] The new element of a “2X” device, in relation to the simple configuration of FIG., [0081] Obviously there is much more to say about the mirror sub-unit [0082] The next and most comprehensive variant on the nominal design of the wide field Hartmann camera depicted in FIG. 4 is what will be referred to as the “4X” wide field Haronann camera, or better, as the full coverage wide field Hartmann camera, depicted in FIG. 7. The guiding principle behind this device is that no photons are lost other than by simple reflection or refraction. The two previous configurations are both forced to truncate (occlude) the field of view along the central axis of each field angle in order to maintain proper segregation of the spots on the electronic detector array [0083] Referring now to FIG. 7, the full coverage configuration is identical to the 2X configuration described previously, up through the 2X beam multiplexer [0084] For the remainder of this disclosure, the full wide field Hartmann design depicted in FIG. 7 will be assumed and utilized as the preferred embodiment, while the simple and 2X configurations will be mentioned as significant alternatives with the same principles involved. This being the case, we turn now to a relatively detailed design of a full wide field Hartmann camera in order to illustrate the application of the design principles outlined above. FIG. 8 has a schematic breakout of the critical specifications for the example telescopic system which will be used throughout the rest of this disclosure. This example design should not at all be considered as an optimized design for a 2.5 meter telescope, it only uses simple numbers and fractional relationships which can best illustrate the principles of design. Those practiced in the art of telescope system design will understand that gross system specifications such as these are only the starting point for detailed engineering refinement. In fact, detailed investigation of these specifications will find that very little engineering tolerance has been included. This was done to further underline the nominal relationships of the basic designs parameters. Also included in the example system design is a few significant specific variations on the “symbolic” outline of FIG. 4 with FIG. 7's full design substituted. These variations illustrate the points made earlier on how specific design scenarios (such as a 2.5 m telescope w/20 arc second square field of view—FOV) can give rise to specific optical engineering design realizations. This point is most evident in the realization of the design of the field tenses [0085]FIG. 8 is best described by starting with the specification of the primary wavelength λ=500 nm. Those practiced in the art will recognize that this is in actuality a specification of the central wavelength of the narrow spectral band light beam which will fall upon the primary detector [0086] Turning now to the full wide field Hartmann camera in FIG. 8, it has been chosen that no magnification optics are used and the system directly uses the F/10 of the primary telescope. This gives a plate scale at the first primary focal plane [0087] This concludes the detailed description of the camera system [0088] The Camera Controlling System and Data Collection System, [0089]FIG. 9 depicts the camera controlling system and the data collection system. The state of commercial technology is such that this entire depicted system is readily available by configuring commercial product(s). Essentially, it is what is commonly known as a digital camera system or a digital imaging system (among many other names as well). It is well known in the practice of systems engineering (and computer system configuration) that there are as many ways to arrange these commercially available elements as there are systems designers. The figure attempts to do justice to this multitude by leaving the relationships quite general. The only feature which could be considered somewhat unique to the needs of the illustrated embodiment of this invention are the shutter synchronization sub-system [0090] For the purposes of specificity, the preferred embodiment of the invention utilizes CCD detectors as the focal plane arrays of choice for the visible and near infrared light region. Other cameras operating in the infrared are certainly possible, as well as non-CCD focal plane arrays in the visible/IR. Both CCDs, [0091] Data Processing [0092] To understand the data processing methodology of the preferred embodiment of the invention, it is helpful to first generalize the input data to the processing steps. Physically, in the preferred embodiment of the invention already described, the input data is what is contained in the frame store buffers [0093] 1) A set [0094] 2) A set [0095] 3) A set [0096] The data processing methodology disclosed as the preferred embodiment of this invention in effect operates at this most general level of abstraction. [0097] A processing methodology according to this aspect of the present invention creates a single clear image out of a set of E field variant distorted digital images (or digitized analog images), a set of data providing “direct or derivable” information about the field variant distortion itself, and any other a priori knowledge of the system design or the “statistics” of the distortion itself. [0098] As so defined, this processing methodology is applicable to any linear imaging problem affected by field variant distortions and capable of some form of quantifying that distortion and its field variance. It operates on a set of distorted images to create a single clear image. This end result is analogous to a machine which would take several exposed pieces of “blurred negatives” of an object and produce a single “unblurred negative” of that object. An image deblurring system and methodology, per se. The large dashed boxes in FIG. 10 surrounding the input [0099] Digression: General Mathematical Background [0100] Before describing the details of the flow diagram in FIG. 10, and the ensuing further details of the specific preferred embodiment contained in later figures, an extended digression into a useful mathematical language is first in order. This is necessary if for no other reason that complicated design considerations can be expressed in concise, but unequivocal terms. A much more thorough discussion of the general mathematical considerations involved in the basic problem of image restoration can be found in chapter 2. “Two-Dimensional Transforms” written by H. C. Andrews, contained in volume 6 of “Topics in Applied Physics: Picture Processing and Digital Filtering,” ed. T. S. Huang, Springer Verlag 1979 or in chapter 1. “Signal Restoration, Functional Analy sis, and Fredholm Integral Equations of the First Kind” written by C. K. Rushforth, contained in “Image recovery: Theory and Application,” ed. Henry Stark, Academic Press. 1987, both of which are available in the libraries of large technical universities. There are several other such sources as well. Rather than continually referencing such sources, I will focus in on those items pertinent to this discovery in the following discussion. [0101] A helpful perspective to view the processing methodology is through what is commonly known as the linear theory of imaging systems (quasi-monochromatic, unless otherwise indicated). It is summarized in the broadly accepted mathematical expression: [0102] In the general case of this invention where a multiple set of E digital images is collected, it is best represented as: [0103] Here, f is a vector which represents the unknown object being imaged, g is the set (vector) of digital data collected by a digital imaging system, and H is the matrix which describes the linear relationship between each data point in the vector g to each point in the vector f. In actual practice, true linearity is never absolutely achieved, but it is well known that the residual deviation between “linear assumptions” and “non-linear actualities” can easily fall within the scope of a “system error budget.” [0104] For so-called “space invariant” imaging systems, i.e. field invariant systems as this disclosure refers to them, each point in the vector g has an identical relationship with the set of points in the vector f, except for a shift in indices. Expressed in terms of a digital camera, this implies that one pixel has a certain “view on the object space” while its neighbor has the same view, only positionally shifted [i.e FIG. 3 where the window function does not change shape as it moves from one data point to the next.] Many textbooks exist which explore these basic points in much greater detail. The point here is that since so many practical imaging systems come reasonably close to this “space invariant” approximation, much of current image processing techniques revolve around this assumption since it affords a wide variety of mathematical and computational simplification. An enormous amount of simplification, usually. The deviations of an actual imaging system from this approximation are most often small enough so that the errors caused by the approximation are well within system error budgets (the same as with the “linear approximation” mentioned above). One of the most broadly accepted and widely used manners in which this field invariant approximation is expressed as a mathematical relationship is: [0105] or [0106] These look quite similar to equation 1, but have indeed undergone a major simplification. Equation 2 is what is known as a convolution equation (as symbolized by the convolution operator {circle over (x)}) and equation 3 is its equivalent expression in fourier space, where the capital letters symbolize the fourier transform of the lower case space functions g, h, and f of equation 2. These are all common expressions in the mathematical representation of physical systems and are widely recognized and utilized by the engineering community (again, reference the texts cited above). Even though ultimately ALL imaging systems are FIELD VARIANT, so many systems are close enough to field invariance that the errors introduced by using equations 2 and 3 in practical applications are negligible. [0107] Not so with the problem of imaging objects through a turbulent media such as the atmosphere. however. That is, IF one wishes to image objects which are not merely confined to the “isoplanatic region” as previously defined. Even when an application calls for the imaging of an object confined to the isoplanatic region, it will be found by anyone experimenting with the prior art techniques, in relation to the techniques of this invention, that the “isoplanatic assumption” leads to higher noise levels (per number of images collected) than this invention. [0108] To use the common vernacular of the digital image processing industry, the overall problem of imaging through a time varying field variant distorting media can be defined as: “Estimate f given a set of E digital images g [0109] The very definition of isoplanaticism incorporates the idea that h of equation 2 does not change other than a shift. As already mentioned, the transfer function of the atmosphere, i.e. the rows OR columns of H in equation 1, changes quite appreciably across the angles of the sky. So much so that even with a shift of angle in the sky of 10 arc seconds of one degree, in normal “good” seeing conditions commonly referred to as r [0110] A major foundational novelty of this invention (not here “claimed” per se, but leading to specific methodology inventions) is that the jump from equation 1 to equations 2 & 3 is emphatically not made, and that other approximations are introduced (which MUST be introduced IF the enormous computational requirements are to be accommodated by less than all the computational facilities on the planet put together) at appropriate stages, all leading to specific processing methodologies which form the heart of the so-called “image deblurring machine” of FIG. 10. [0111] Instead, the problem is reformulated into the following: [0112] Equation 4 is nothing more than a restatement of equation 1 (barring the subtleties of the continuous-discrete transformation process, here assumed to be “one and the same”) with the additional consideration—normally quite helpful in actual instrumentation—that the coupled variables (x-y) in h are simpler than x and y by themselves. This is to say, the matrix H has a roughly bandlike nature (the word “roughly” is used owing to field variance and geometric distortion) and use of the coupled variables allows for a simpler conception and description of h. The careful observer will note that the “known” statistical properties of a turbulent atmosphere do play a role in this decision to utilize the more intuitive coupled variables (x-y), in that these statistics indicate a low spatial frequency of the “change in h [0113] Equation 5 is the same as that which is depicted in FIG. 3, implicitly including the atmospheric contribution and the telescope optics inside the field variant h function. Even the pixel sensitivity function, p [0114] This is nearly a restatement of equation 4 in a discrete format. The vector t[n] (as a convenience) is a positional vector about which the h function can be defined. That is, it preserves the general convolutional character of the problem. Moreover, it can accommodate lower order geometrical and/or “center-of-gravity” “warps” if any advanced designs and methodologies require such, yet retain simpler mathematical expressions. [0115] At this point, a kind of axiom is needed which states that much of the subtlety contained in equation 5, and contained in the use of the vector t[n] of equation 6, is within the proper scope of advanced engineering and CAN be ignored while describing the illustrated form of the invention itself. The “errors” which will be introduced into the overall system performance will remain just that, errors within an established system error budget. This is common engineering practice to use approximations and let the errors fall out where they may. It is the nature of the approximations chosen which will determine whether these errors are within acceptable limits. So with this axiom in mind, the following approximation is put forth which can get us back on the explanatory path of the preferred embodiment of the invention. I return to a continuous function representation of the problem where the pixel sensitivity function will now be considered to be incorporated inside the h function: [0116] This is precisely equation 4 only now h tacitly includes a field invariant pixel sensitivity function, p(x). From here, we can follow a path which will wind up with an equation that will inform many of the engineering choices made in the preferred embodiment of the invention. [0117] Focussing our attention first on the continuous function h, one well known manner of approximating (arbitrarily closely) h is: [0118] This leads to interesting patterns and insights but not the ones that the preferred embodiment is after. Instead, the preferred embodiment can be better appreciated by employing the following expression which can generally be made arbitrarily precise: [0119] This has been cited in the general literature as a “series expansion of separable functions,” see Andrews cited above. I am currently unaware of efforts which have gone beyond merely citing the more general form of equation 8 and noting its lack of usefulness in a global restoration context. Here, used with the (qs-y) variable on the z function, careful examination will show that this is merely a general interpolation function, z, being applied to a “table” of h functions where the table entry spacing is s. The interpolation function employed is left as an engineering exercise, though the assumption of a bi-linear function is a reasonable, practical starting point for this discussion [spline functions, standard square box functions, truncated cosine squared functions, etc., are all candidates with various properties] The main purpose of equation 8 is the separation of the variables y and (x-y), mixed together in the original h function, into two functions, and in a way that retains a simple spatial coordinate system for the field variance (which the grid of table functions with a spacing s provides). The in-operation “typical” frequency spectrums of these two functions will grossly mimic the “field variance of h” in z, and the “intrinsic local transfer function of h” in h itself. It should be pointed out that, in the utmost efficiency minded approach to the design of a system, equation 8 may be somewhat wasteful if the spatial frequencies of the various basis components of b differ markedly. This is to say, the pole spacing, s, of the interpolants will be dictated by the highest spatial frequency components of h, and may be “overkill” for other components that would have sufficed with a pole spacing longer than s. It is felt that this is of negligible concern at this point and that the matter is entirely left to engineering if the phenomenon truly causes undue inefficiencies. A double summation, separable equation 7 could then be employed if absolutely necessary. Plugging this new representation of h into equation 4A, we get: [0120] giving a fourier counterpart: [0121] where the shifting, qs, in z is now merely indicated by a subscript q on Z. In general, equations 9 and 10 are no different than equation 1. However, there is now a certain separation of the “localized” (and generally lower frequency) components of the field variance from the more familiar transfer characteristics symbolized by the local H function indexed by q. Careful examination of equation 10 will show that in effect, this is the fourier counterpart to equation 1: [0122] where the matrix, H, is in italics to differentiate it from in space domain counterpart, H. For the sake of computational simplicity and conciseness, it would be helpful to quickly outline the detailed relationship between the F and G. Expanding F into a series expansion and pulling the summation out front, equation 10 turns into: [0123] where it can be seen that everything inside the outer parentheses, including the known and/or fully derivable H.sub.q functions, is now a single knowvn function: [0124] Once again, this is merely reformulating equations 10 and 11, emphasizing that the rows and columns of the H matrix in equation 11 can be easily calculated on a computer. Detailed examination of equation 10 through 13 will find that the H matrix is also roughly banded just as its space domain counterpart H in equation 1. This concludes the general “abstract” mathematical discussion which will be referred to in later sections describing the preferred embodiment in detail. [0125] Turning now, to a highly practical mathematical formalism in which to further approach the computational necessities of the preferred embodiment, refer back to FIG. 3. The function h in equation 4A is graphically depicted in FIG. 3, where two particular points in g have their own unique h functions projected onto object space. It should be noted that each and every point in g has a unique projection onto (window on) f. FIG. 3 also depicts a variety of steps taken which begin to transform equation 4A into the preferred embodiment of the invention. The main step involves choosing a set of orthonormal basis functions over which to define f (and subsequently h). It is well known in the engineering community that there are many such sets to choose from, each having differing properties for specific applications, though all serving as a sufficient mathematical description of the functions. The choice of basis functions for the preferred embodiment of the invention is anchored by a uniformly spaced grid whose linear extents along the two axes correspond precisely to the dimensions of the pixels on the primary array [0126] Now there are a few practical issues which arise at this point which a meticulous engineer can notice and which, before being swept under the rug as prima facie negligible, should at least be highlighted and analyzed. It is pertinent to the invention, and pertinent to a full understanding of how to implement an embodiment of the invention, to list these issues and discuss them briefly in the context of building an embodiment according to the present invention. There may arise situations where this discussion will be highly pertinent to engineering details, especially in systems designed to image through extremely turbulent conditions. [0127] To begin with, the use of any finite set of basis functions to represent some object [0128] Continuing then, there is the classic and widely utilized theory of the telescope which culminates in the concept of the diffraction limit. This theory, in a rough summary, postulates the existence of a finite aperture viewing an infinite extent object at an infinite distance in some particular wavelength of the electromagnetic spectrum; Using a fourier mathematical representation now, any point P inside an enclosed instrument will have a window function h (of equation 4) which is absolutely limited in (angular) frequency to a period of arc tan(λ/A), where λ is the wavelength of light and A is the maximum aperture distance, both measured in the same units. The driving force behind this theory of the diffraction limit is the idea that, as a luminous point source is moved around in the object plane, the complex phase of the incoming planar light waves as projected onto the “pupil” (the open area of the aperture) can only relatively modulate at this maximum frequency and no higher. Any point inside the instrument will receive a “null signal” from angular frequencies higher than this “diffraction limit” angular frequency with the period as stated above. In broadly accepted theory, only the cosine multiplicative component on both the complex amplitude and the complex phase of the incoming light waves creates frequencies above the diffraction limit, where these two components are vanishingly small at the diffraction limit frequency even for very small aperture imaging system (they at least, however, provide a solid mathematical negation to the idea of an absolute diffraction limit and to an absolute null space). As a final note on describing conventional theory and its telescope design implications, focal plane arrays are generally chosen to have a spatial pixel spacing which provides 2 pixels per cycle of this maximum angular frequency, though some applications will specify a greater or lesser spacing for a variety of reasons. Returning now to the relevance of all this to the current invention, it can be noted that the conventional theory of the diffraction limit posits the existence of planar light waves impinging at the pupil [0129] Some general engineering observations are in order at this point. The overall point to the above discussion is that the pixel spacing may need to be somewhat smaller than would otherwise be the case for a straightforward diffraction limited calculation. Furthermore, an object estimate MAY be sought which exceeds the frequencies of the so-called diffraction limit. Having said all of this, it will be found that for large telescopes in particular, where the spatial bandwidth of the field variance as reflected in V is much smaller than the spatial bandwidth reflected in H, the effect can be considered negligible. On the other hand, in smaller aperture telescopes where V begins to appreciably approach H's spatial bandwidth, the effect can be significant to the design. As a very crude rule of thumb, the pixel spacing should be decreased from that dictated by normal diffraction considerations by the amount: sqrt( [0130] where p is the pixel spacing in arcseconds dictated by traditional diffraction theory and s is the “table spacing” of the h functions in equation 8. It can be seen that for large aperture telescopes this factor is quite negligible in even some of the most severe turbulence conditions. Be this as it may, the effect is one to take note of, possibly, in non-conventional imaging system design applications. [0131] The above discussion overlaps to no small degree with the issue of aliasing previously mentioned. The “safe” route is to pick a pixel spacing conservatively above the modified diffraction limit indicated in equation 14. However, this can lead to less photons per pixel and possibly higher noise levels in certain critical system results specifications. In these cases, often a choice is made to design a spacing which is some significant spacing larger than is dictated by equation 14, improving the S/N ratio at the expense of “tolerable” aliasing. As always, the choice of exact pixel spacing, and the trade-off between noise and aliasing, is best left to specific engineering refinement. [0132] This concludes the general mathematical description of the basic problem and the discussions on some of the mathematical subtleties that may be encountered. [0133] The Preferred Embodiment of the Data Processing Mehtodology [0134] Returning back to FIG. 10, we can now overview this processing methodology aspect of the preferred form of the invention. We can then quickly transition to the specifics of the preferred embodiment. The following discussion is only meant as an explanatory discussion of how the major pieces fit together, while the recipe for the prefcrred embodiment will come later. [0135] The data input to the process, [0136] Getting back to FIG. 10, then, the heart of the processing methodology is the loops indicated by steps [0137] Now that the steps have been generally described, it is easier to describe one novelty of the processing methodology aspect of the present invention. To sum it up in one sentence, the illustrated embodiment of the invention simply performs a weighted average inversion—“reversion” as dubbed herein—defined over a generally overdetermined set of signal to noise ratio segregating basis functions. In plain English, the process not only “undistorts” the individual images, it does so in a manner that allows for all images to be properly averaged together into one final crisp image. An individual “undistorted” image IS generally noisy, and its noise properties non-uniformly distributed. Moreover, this non-uniformity of noise properties changes significantly from one image to the next. The present process, through the use of what could also be called universal signal to noise ratio preserving basis functions, creates a common ground for the image to image noise property differences to be reconciled, while the weighted average “reversion” not only ensures a maximally efficient signal to noise ratio on the final image estimate, it hugely reduces the practical difficulties associated with divisions by very small or null entities. Null entity inversion at the final step [0138] As a final mathematical note before the detailed recipe for the preferred embodiment is described, we can further analyze why the universal set of basis functions defined in step Φ [0139] where c*c*d*d=MN (this is cryptic, admittedly, but the words should speak for themselves and it is quite immaterial to the disclosed invention). It should also be noted that in the case where the overall imaging system can be well approximated by the isoplaiatic assumption, i.e. equations 2 & 3 can be profitably used, then it can easily be seen that the sinusoidal basis functions of the standard fourier components can fully crve as the “universl” set of signal to noise segregating basis functions. Indeed, this very concept is superfluous in that case; field variance is what gives it real meaning. It should also be noted that the prior art, most notably in the paper “Deconvolution from wave-front sensing: a new technique for compensating turbulence-degraded images,” J. Primot et. al., J. Opt. Soc. Am. A Vol 7 No9, p 1598, not only implicitly makes use of this clean segregation of the signal to noise components in the isoplanatic case, its use of the time average of the correlation by the time average of the auto-correlation, equation 3, is a degenerate case of the general weighted average reversion listed above, i.e. it happens to degenerate into global orthonormal basis functions which are derivable through less general operations such as correlations. It is clear, however, that the correlation and autocorrelation operations lose their efficacies when applied to field variant systems and thus do not extrapolate beyond the “isoplanatic” assumption. In any event, this functional congruence of this equation 3 to the more general principles outlined in FIGS. 11 and 11A dictate that this invention disclosure will not presume to claim the use of this specific processing technique within “fully isoplanatic” methodology. On the other hand, “fully isoplanatic” applications which utilize “field variant technique enhancements” certainly fall under the purview of this invention and will be claimed accordingly. [0140] The Detailed Preferred Embodiment [0141] The description of the preferred embodiment picks up where it left off with the detailed descriptions of the camera system [0142]FIG. 12 contains a schematic of the information which is best described as belonging to the input system data [0143] Referring back to FIG. 10, the first bona fide processing step is the “non-computational” definition of the universal S/N segregating basis function set. FIG. 13 schematically describes the set chosen for the specific example described in FIG. 8. First, the digital image is split up—using square spatial weight functions—into a spatial grid of “data sub-zones” [0144] First of all, there are clearly choices which could be made which will greatly reduce the overdetermination factor. The example presented in FIG. 13 chooses a conservative approach more appropriate to empirically derived error budgets as opposed to well-defined theoretical ones. The thinking behind the choice of a 32 by 32 region is that it is “well within” the “isoplanatic” conditions such that the pixels within the region can be treated in a normal isoplanatic processing methodology. Likewise, it is not too small as to unduly increase the computational time and resources in the hunt for diminishing returns (negligible decreases in error and noise). It is clear that the ideal universal set of basis functions would be to have each pixel define its own 1K by 1K UV plane, giving 10 [0145] Returning now to FIG. 13, both the choice of a 32 by 32 sub-zone, and the choice of a 256 by 256 “halo” [0146] To finish the description of step [0147] Turning now to FIG. 14, we find the full and specific process steps—or flowchart—for the software which carries out the standard operations depicted in FIG. 10. As with the case of the specific embodiment of the wide field Harnnann design in FIG. 8 in relation to the general design in FIG. 7, the flowchart in FIG. 14 will make specific choices of options outlined generally in FIG. 10. Prototype software and general implementation strategies have made use of the common software language known as “C,” and therefore much of the presentation of the flowcharts conform to the basic concepts employed within that programmning language. In particular, the use of callable “sub-routines” are used which then call forth other figures and descriptions, as opposed to describing these sub-routines in absolute linear execution format. The use of sub-routines also tremendously eases the discussions on the variants of subroutines, since the discussion on variants can be confined to those appropriate sections and not overly burden the description of the main program itself. [0148] The preferred embodiment comprises a computer program written, debugged and compiled which shall be called “main” as is quite archetypical to C language computer programs. Its is well known within the profession of software engineering that such a program as this could easily be a sub-routine inside a larger program, such as graphical to text-based user interface program or a larger project organization program. Apart from whatever normal software engineeting considerations which enter into the creation of computer programs, the first programming step [0149] These three arrays can be interpreted as a 32 by 32 array of 128 by 256 UV plane elements: that is, the real numerator 356, the “imaginary” numerator [0150] The next step is what is commonly known as a nested loop or an inner loop [0151] Next, step [0152] A third “nested loop” [0153] After step [0154] The next step is to enter a loop [0155] It can be seen in FIG. 15B, and noted by step [0156] Finishing up with the main program and FIG. 14, the loop [0157] The Wavefront Estimation Sub-routines [0158] There are two reasons that these sub-routines are split out into their own section. First, they are generally quite specific to a Hartmann type wavefront camera in particular and to atmospherically distorted imagery in general (as opposed to the most basic category of “distorted digital imagery”). Second, they are relatively involved and contain their own alternative methodologies (generally based on the typical “accuracy vs. resources” engineering dynamic) and therefore they would have diluted the discussion on the main program to the point of unintelligibility. Now that the rationale for this has been stated, we can turn once again to the specific preferred embodiment which is based on Hartmann “spot” data being the input data [0159]FIG. 17 depicts a simple overview of the wavefront distortion methodology. It is specific to the use of the Hartmann camera as mentioned. It is important to point out that the end product of this entire wavefront section and the described sub-routines is to create the complex array “rev_array” [0160] Of all the three procedures listed in FIG. 17, the calibration procedure is most affected by which type of Hartmann camera system is being employed. The splitting of the calibration into two delegates the bulk of the work to the Macro routine, while the Micro routine “fine tunes” the estimates of the instantaneous spot center references. Three particular choices of Hartmann camera system design were detailed earlier in FIGS. 4, 6, and [0161] 1) find which field angle it belongs to, and to which nominal area in the pupil plane does it correspond. [0162] 2) find its nominal “averaged” center location (or zero-slope reference) as a function of the pixel grid on CCD [0163] 3) determine if its nominal center point on the pupil plane is too close to, or over, the edge of the pupil support, thereby precluding a spot from being considered a “good spot”; that is to say, its inherent sensitivity will fall below some predetermined threshold. [0164] Additionally, each field angle in the array of sampled field angles needs to be matched to particular location on the primary CCD [0165] Using these three tasks as guidelines, and the need for positional correlation as the parallel companion, much of the macro calibration belongs to the realm of common sense and basic engineering. There is obviously a great deal of regularity in the arrangement of the spots. Contrarily, there may also be subtle deviations from this regularity which, especially in the case of the full Hartmann camera system of FIG. 7, would require a rather involved software scheme for “keeping track of the spots.” It should be pointed out that prior art (paraxial) Hartmann sensors already must deal with, and have already come up with fine solutions for, a somewhat simpler form of macro calibration. It is “somewhat simpler” due to the fact that there is only one field angle involved, where in the wide field case there are interlaced field angles—multiply interlaced in the 2X and full designs. Referring back to FIG. 4B, it contains the simplified version of the interlaced field angles for a 4 by 4 field angle sampling Hartmann camera system (simple configuration). The idea of spots “fading out” as opposed to abruptly stopping is roughly drawn in as well, suggesting the need to threshold the “good spots” from the bad ones. [Again, parasial prior art already has developed solutions for this type of problem. Also, yet another reminder should be made that the clear circular fade out in FIG. 4B would actually be more of a fade out of the squares as a unit, not the individual spots within the 4 by 4 sub-groups.] [0166] The preceding discussion and FIG. 4B are quite sufficient to map out specific software embodiments. Be this as it may, a description of further considerations is in order, and detailed descriptions of an “inefficient but thorough” output file will be presented in order to mesh sufficiently with the discussion on the micro calibration and the two remaining procedures depicted in FIG. 17. [0167] It might be possible to design the Hartmann camera system with such tight opto-mechanical tolerances that much of the macro calibration could be performed “by design.” This is to say, the physical tolerance would unambiguously assign spots to field angles and pupil positions, and field angles to primary CCD (8) pixel locations. This would generally involve micron level tolerances for some components, such as the CCD relative to the lenslet array, and slightly eased tolerance on other components. It would be yet more difficult to accomplish this for the 2X and the full models. It is equally probable that this kind of ultra-precision—for this particular “need” at least—is not necessary. What is more needed is in-operation rigidity rather than absolute accuracy. Even this need for “rigidity” is contingent on whether or not “engineering methods” such as the micro calibration are available and economical. In any event, one extremely simple method for assigning spots to field angles (as a first task) is to “scan” a bright star in and around the paraxial field angle of the telescope. Those experienced in the art of telescope operation will understand that, assuming we gather nominal digital exposures, the bright star will variously “light up” the interlaced arrays of spots belonging to the individual field angle in which the star currently is positioned. Provided that the bright star is positioned such that it is sufficiently centered in a given field angle, those spots belonging to that field angle readily manifest themselves. If it is not centered, either no spots light up, or two “sets” of spots will light up such as in the full model of the wide field Hartmann design, when the star falls at the border of two (or four!) regions. Once a bright star unambiguously lights up a single array of interlaced spots, standard prior art methodologies can treat those spots as if they were from a normal paraxial Hartmann sensor, find their nominal centers (over an ensemble of exposures), find their “ [0168] During the above “scanning of a bright star,” the primary CCD [0169] Let us now clearly define the required output of the macro calibration routine, realizing that the above discussion is sufficient to outline how this output is generated. The use of the “C” struct variable type is enormously helpful in this context. The output can be seen in its most thorough (but probably inefficient) form as:
[0170] The reason that this is “inefficient” is that no allowance is made for the regularities involved, regularities which could be exploited for reduction in storage and computation time. Nevertheless, this disclosure shall use this as the foundational approach. The output is arranged as an 8 by 8 array of structures, each with the associated elements inside the brackets. Each field angle has a nominal center point on the primary CCD stored in the first floating point values listed. The second line contains one integer value which gives how many “good spots” were found for this particular field angle. Note that this allows for each field angle to have a slightly different number of good spots than the ensemble average; i.e., field angle X may have 120 good spots, while field angle Y has 122 good spots. The third line contains the x and y coordinates of the nominal “ensemble average” center location of all the good spots for this field angle. It has “num_spots” number of floating point values referenced to integer pixel locations on the CCD [0171]FIG. 18 takes a closer look at the situation obtaining for the specific preferred embodiment. It can be noticed that the top drawing in FIG. 18 corresponds to the appearance of a digital image deriving from the full or “4X” beam splitting Hartmann camera of FIG. 7. In particular, it depicts the (nominally optimized) configuration as would appear for the 2.5 meter telescope, utilizing a 2048 by 2048 CCD array as the specific focal plane array [0172] We can now turn to discussing the second procedure, “step 2,” in FIG. 17, the “global wavefront sub-routine” [0173]FIG. 19 can help better explain how these general requirements are implemented. The first step [0174] Whether step [0175] Steps [0176] Loop [0177] The third appointed task, that of further processing the array of point spread functions [0178] The best place to begin describing the preferred embodiments of the generation of the specific rev_array [0179] As an important but ancillary mathematical digression, solutions [0180] I will first describe the simplest and fastest reversion methodology marked as solution [0181] At this point it is worthwhile to compare this most simple reversion solution (number [0182]FIG. 22 contains a schematic series of methodologies which, moving from the top drawing to the bottom, transition from the “obvious step and repeat” method of isoplanatic prior art processing to the preferred embodiment solution [0183] Stepping out now from extension of the prior art techniques into the novel design of processing methodologies which borrow on these prior art techniques, the second schematic in FIG. 22 takes a jump downward in error performance from number [0184] We can turn now to a description of the preferred embodiment of the processing methodology; specifically, the use of the data point window function based reversion method, labelled as solution [0185] The output of the sub-routine [0186] The remainder of processing steps in FIG. 23 translate the concepts in the two drawings of FIGS. 24A&B into computational routines. The next step after allocating the large array bilin[ [0187] [Efficient software engineering may wish to use quad or octal symmetry to merely store and use a smaller segment; but the computational overhead in this routine is so small that using the full array can make for easier pointer arithmetic.] [0188] Step [0189] This completes the detailed description of the preferred embodiment, the one which uses the data point window function as the basis for the reversion methodology. I turn now to a discussion of what is the broader de facto preferred embodiment of the invention. It is “de facto,” since both of the preceding solutions [0190] Recalling the discussion surrounding equations 10 through 13, and turning to the drawings in FIG. 25, we can outline the technique for refining the specific reversion methodology to be used. The top drawing in FIG. 25 presents a simplified view of a data point centered at the origin for convenience, having the point spread function of the nominal field angle of that data point drawn an x with a circle around it directly at the origin as well. Likewise, there are field angles nearby which have slightly different point spread functions, where the distance scale here might be spaced 1″ apart, somewhere within its isoplanatic region which would isolate the “low order” variability in the point spread function. Those practiced in the art will realize that there is still a high degree of correlation between the point spread functions at these nine locations, but in some sense there is a local “tip-tilt and low order curvature” present in the field variant point spread function. In other words, if we focus in on only the local low order change in the point spread function, this will be closely approximate to the actual characteristics of the “change” in the point spread function in the neighborhood of the field angle at the origin. We can therefore imagine that modelling “low order” change in the point spread function in this small neighborhood is better than modelling no change at all (as in the case of solution [0191] Finally, solution [0192] Advanced Wavefront Estimation Routines [0193] Whereas the processing methodology presented in FIG. 19 and discussed in the disclosure text was quite functional and “enabling,” it cannot be said to be the “best mode” of the invention. The only reason for this is that it tends to have higher residual wavefront error than the routines which follow. Referring back to FIG. 5, there is box [0194]FIG. 26A gives a closer look at what is taking place inside the wide field Hartmann camera system. A bright star [0195]FIG. 26B sets up some generic mathematical concepts and a framework which can be used in creating computational solutions to improving the wavefront measurement/estimation. The Full or 4X wide field Hartmann design of FIG. 7 and FIG. 8 will be the basic model used in the drawing, where the 3.6″ field angle spacing of the specific preferred embodiment is also used. The top left drawing in FIG. 16B is a three dimensional array representing the “current” multi-spectral object estimate 610 corresponding to one of the field angles in the 8 by 8 array of sampled field angles. It is a course resolution estimate of 1″ square boxes of 4″ square total extent, giving a 4 by 4 pixel estimate. The double summation below merely tracks the coefficients of the values and leaves off the implied square box basis functions, as is common. The spectral plane can be considered as 20 nanometer square bandpass regions corresponding generally to the spectral regions of significant energy as defined by the rightmost spectral curves of FIG. 12A, the wavefront beam sensitivity function. If a monochromatic “artificial beacon” is the only signal of significant energy content, then there will only need to be one spectral plane rather than several, where this beacon assisted system is also depicted in FIG. 12A—effectively reducing it to a 2 dimensional object estimate. The coarse [0196] Among the many choices of better estimating the true pupil function, the preferred embodiment of the invention chooses a rather simple approximate scheme loosely based on what has generally been known as the Van Cittert solution to deconvolution problems. [The basic methods set down by Herr Van Cittert in the late twenties and early thirties have gone through the nth degree of refinement and modification. Nevertheless, using the “Van Cittert” label still carries the basic philosophy of solution with it.] It is a little bit ironic that this disclosure chooses such a solution, since much of the philosophy of the invention as a whole is to boil down certain relationships into direct (quasi-deterministic) forms rather than the more indirect Van Cittert formalism. Be this as it may, the Van Cittert solution will indeed effect improvements at a reasonable computational cost, and future engineering refinement will be tasked with finding better quasi-deterministic methodologies which are better candidates—as if they were plotted on the same type graph of FIG. 20. [0197]FIG. 27 has a general processing flowchart of a new sub-routine which should be placed immediately following step [0198] Again, it should be clear to those practiced in the art of image processing and in general “data correction methodologies” that the example processing methodology of FIG. 27 is indeed indicative of an entire genre, iterative and quasi-deterministic alike. Entire books have been written on the basic subject and those building on the present work are encouraged to consult those books and apply the basic solutions to the specific conditions at hand. With this said, we can turn to a brief discussion of FIG. 28 wherein a rough refinement feedback system is depicted on the processing methodology of the invention as a whole. At its core, this could be conceived of as a grand web of Van Cittert like feedback loops which in principle can iterate its way to a refined self-consistent solution for an object under study. However, I believe this interpretation overstates the argument for the very need for these refinements, and in situ testing of actual telescopic systems will be the final arbiter of whether such loops are employed. Furthermore, the more that such “feedback loops” are replaced by direct, efficient prediction methods, the better. In some senses, the Van Cinert approach is hence a fail safe when there is appreciable error which can be removed, but the quasi-deterministic solutions are either computationally impractical or they are not within the grasp of the imagination. In any event, some or all of the loops depicted in FIG. 28 should be considered as a potential part of a working model of the invention. I personally like to envision them as single iteration loops, since the errors they are correcting are already very small (except in the case of FIG. 27 perhaps), and one iteration will bring them well within the point of diminishing returns. The elements of FIG. 28 that should be particularly noted here are first the step labelled “spot cross-talk correction” [0199] A few additional refinement possibilities which have or haven't already been mentioned and which did not easily fit into FIG. 28 are those connected with improving the generation of “rev_array” to the ideal of the full matrix solution of equation 14 (i.e., yet more discussion on the search for solution [0200] Concerning polarization effects and their potential introduction of further error sources, this is an issue present in this invention and in the prior art systems. Once the basic solution has been formulated in the manner of equation 1, the isolated issue of polarization becomes one of predicting the rows and/or columns of H within this additional “polarization” context. The issue of how the “spot information” on the Hartmann data is different from the “true pupil function” and the focal plane array [0201] One final note is on the broad issue of tracking an object as the earth rotates. Though there are volumes which can be written about this and its effects on the methods of this invention, perhaps the easiest summary of the effects of “tracking errors” is that they should, in fact, blur the final image in the same manner that they have always done. At first glance, this appears to place major new demands on the tracking abilities of large telescopes, suggesting they go from there current 0.1″ region specifications to an order of magnitude better (I can hear opto-mechanical engineers groaning). Not to worry however, there are several interesting tools which can be brought to bear on alleviating this problem, POSSIBLY TO THE EXTENT WHERE TRACKING SPECIFICATONS CAN ACTUALLY RELAX. Wouldn't that be a twist. The basic idea is that a phase only corrected version of a speckle image should have stable overall “centroid” properties, and that by either taking the individual exposure numerators RE, IM. [0202] This concludes the disclosure of the preferred embodiment of the invention. The following sections describe major variations or extensions thereof applied to systems which are different enough to deserve being highlighted. [0203] Alternative: No Wide Field Wavefront Distortion Information Available [0204] Modified methods of the invention can be applied to physical systems and to sets of blurred digital images wherein no direct information is available concerning the distortion of the images. In effect, the situation would be such that there would be no input information of the type [0205] Getting back to the issue at hand, the case we examine now is when there is no input data set [0206] Special Section: The Use of Artificial Beacons [0207] It has long been appreciated that wavefront sensors generally require a moderate incoming light flux in order to operate properly. For the case of atmospheric turbulence, their short exposure times increase the need for higher light levels. Both theoretical limits and practical state of the art instrumentation limits have generally limited conventional wavefront sensors to the 10th stellar magnitude object brightness or perhaps a little better. [It is actually dependent on a variety of factors.] Within the last decade or so, practical systems have been built which provide additional light levels to the wavefront sensors via, generally, laser light directed along a near-paraxial field angle of the telescope; the laser light either backscatters back into the telescope, or may instead excite florescence in higher altitude atmospheric layers with a small portion of the energy once again going back to the telescope. These systems now go under the name of “artificial beacons” generally. There ultimate purpose is to extend the faintness limit of objects which can be imaged by the telescope (almost exclusively in the context of adaptive optics systems; see the discussion on prior art). These artificial beacon concepts can also be quite helpful in conjunction with the techniques of this invention. [0208] The wide field nature of the preferred embodiment of this invention dictates that artificial beacons should cover the field of view desired, as opposed to ‘some close location to an object of interest within the isoplanatic region’, as is often currently the case. This suggests a 64-fold increase in optical power (not a trivial matter) from the source of the artificial beacon over and above that required for the isoplanatic case, merely deriving from the 8 by 8 isoplanatic regions being covered (and assuming comparable limiting magnitudes of the Hartmann sensors used). We have already seen how the wide field Hartmann sensor integrates its signal from an extended field of view, thus, there is no inherent reason to localize the beacon light into a “star.” If particular applications could benefit from such localization, so be it. The point is that for wide field uses of artificial beacons, a very wide artificial beam can be generated and projected out along the field of view of the detector; there is no need to get any fancier. Suffice it to say that the techniques and methods of this invention can certainly make use of this latest methodology of “artificial beacons.” [0209] Special Section: Non-Imaging Uses of the Wide Field Wavefront Sensing System [0210] It is well known in the optical industry that wavefront sensing systems can be employed as central or peripheral elements in non-imaging applications. It is clear that the wide field wavefront sensing system disclosed herein is no exception. Specifically, the disclosed system could certainly be utilized in applications such as single or multi-optical surface metrology (profiling, etc.), distorting media tomography, electromagnetic communications systems including laser-based communications through a distorting media such as the atmosphere, and so on. Overall, it is quite likely that the informational basis of anisoplanatic wavefronts, as measured by and provided by the disclosed systems, will offer significant performance increases, or outright enablement where no enablement was possible before, of many of these non-imaging uses of wavefront sensing systems. [0211] The Wide Field Hartmann Shack Wavefront Camera [0212] This section describes an improvement to the full field wavefront sensor described above. It shall simply be referred to as the Wide Field Hartmann Shack (Wavefront) Camera. It has a less complicated hardware and optical layout, it is simpler on a conceptual plane leading to a simpler approach to the processing methodology, and it should be able to function on dimmer objects. As was the case with the above-described full field wavefront sensor, this wide field Hartmann Shack (HS) camera will function on extended objects as well as on arbitrary contrast extended objects (including zero-contrast “blank walls”). The only potential drawback of the wide field H/S camera is that it may have slightly less immunity to non-singular solutions in very high turbulence conditions relative to the beam-splitting full field design; that is, there may be less tools available to “disentangle” overlapped and inverted signals on the wavefront distortion. This latter potential drawback may be quite minor and may only be of consequence in extremely high turbulence applications not typically encountered in astronomical situations. [0213] The wide field Hartmann-Shack camera is an extension to the general philosophy of the traditional Hartmann-Shack wavefront sensor. Whereas the traditional H/S sensor attempts to measure the local tip and tilt of either a point source or the barycentric shift of a compact diffuse source, the wide field Hartmann-Shack camera described herein attempts to measure the geometrical distortion of an arbitrary extended source of arbitrary dimensions. A point source and a compact diffuse source are special cases of the latter; therefore this wide field Hartmann Shack camera is fully “downward compatible” to single point sources. Another difference between the traditional H/S and this wide field H/S is that the former searches for a mono-axial measurement of tip-tilt centered upon the barycenter of the beacon object, whereas the wide field H/S searches for the field variant tip-tilt across an 2 dimensional array of field angles. Given an undistorted image of an arbitrary beacon object, and a distorted image of the same, a methodology is outlined whereby the distortion is mensurated and related to the tip-tilt of the local sub-aperture wavefront as a function of field angle. Yet another important advantage of the wide field wavefront camera, both the one outlined in this section and the ones outlined in the previous sections of this disclosure, is that they can function on zero-contrast objects provided that true anisoplanatism exists. Those practiced in the art have loosely referred to this as the “blank wall” problem, or the “zero contrast extended object” problem. Both the full field wavefront camera of FIG. 7 and FIGS. [0214] Definite optical considerations not necessary in the traditional H/S sensor come into play when performing detailed design of the wide field H/S camera. This disclosure points out and describes these extra considerations. After discussing these new considerations, the wholly different methodology of determining geometrical distortion will be outlined and described in detail, then followed by a description of how the measured geometrical distortion is transformed into a 2D array of pupil function estimates. [0215] New Optical Design Requirements [0216] The general optical design of a wide field H/S camera is very similar to the design of a traditional H/S sensor. In fact, it is exactly like that depicted in FIG. 4 except that the field angle sampling aperture, [0217] Measuring the Field Variant Tip-Tilt on an Extended Beacon [0218] Reading from FIG. 4, each sub-aperture in the lenslet array, [0219] It remains to determine as accurately as possible what the “undistorted beacon image” looks like, akin to the need to find a nominal center or reference center in traditional H/S sensors, and once determining this, to then find the instantaneous distortion of a given exposure from this “reference image”. TComplicating the situation is the anisoplanatic behavior of the amplitude of an incoming wavefront, i.e. the scintillation. We must also be explicit in now this additional effect is measured and dealt with. Once the geometric distortion between a reference image and an instantaneous image is found, the local difference value directly translates into local tip-tilt, and therefore all previous methods of traditional H/S sensors can be applied to obtain a full aperture pupil function for any given field angle which has sufficient beacon photons to produce a reasonable signal to noise ratio. [0220] Determining the Reference Image for Each Sub-Aperture [0221] The first step in the process is to determine a precise reference image of the beacon object contained within the field stop at the nominal focal plane [0222] The reference image must have roughly equal, or slightly more, high spatial frequency signal level as any given distorted exposure. This will help ensure higher signal to noise ratio measurements of the geometric distortion. At the same time, this requirement complicates the derivation of the reference image. Essentially, the reference image is the same image which would result if no distorting mechanism is present, i.e. if no atmosphere were present. There appear to be several approaches which can achieve the goal of optimally determining the reference image. [0223] The preferred method of this disclosure is to design the wavefront sensing optics in such a way that the individual MTF properties of a given lenslet will be worse than the natural seeing conditions of the primary telescope imaging system. In this way, a primary long exposure image of the beacon object can be gathered on the primary detector and low pass filtered to the same MTF of each sub-aperture and resampled onto the unique pixel grid behind each sub-aperture. It remains to take a long term exposure on the wavefront detector [0224] Other methods of determining the reference image certainly may exist or be developed in the future. Even using the long term sub-images themselves is an alternative, but doing so will need some compensation when it comes time to measure the instantaneous geometrical distortion about the reference image. [0225] Measuring the Instantaneous Distortion Between the Reference Image of a Given Sub-Aperture and an Acquired Short Exposure Sub-Aperture Image on Detector [0226] There are two distinct approaches presented for measuring the instantaneous distortion between the reference image and a given “distorted” image of a sub-aperture. No doubt there will be other approaches developed in the future, each with their own merits. It is impossible to say which of those presented here is the preferred embodiment since each has its own merits relative to a cost/performance comparison. The line integral method does argue to be the more analytical and possibly the more trustworthy, whereas the search method has conceptual simplicity and corresponding simplicity of implementation. [0227] Description of the Problem [0228] FIGS. [0229] The problem is to determine, as precisely as possible, the geometrical distortion function (FIG. [0230] Generic Pre-Processing [0231]FIG. 30 contains an illustration of how the image of FIG. 29 might look as seen by the detector [0232] It turns out that measuring both of these distorting mechanisms independently would be of use to the full distortion compensating system. On the other hand, if all we wanted to recover was the pure geometric distortion, it would still be advantageous to separate the two effects. [0233] It is clear that geometric distortion alone does not effect the total brightness of an object. Another way to state this is that the DC value of the fourier transform is unaffected by geometric distortion. On the other hand, it will be found empirically that the scintillation function finds its maximum value, statistically, at the DC value of the fourier transform and falls off rapidly as a function of spatial frequency. The fact that these two effects occupy opposing regions of fourier space will therefore be exploited in the interests of separating the two effects. [0234] Generally speaking, simple methods of separating the two can merely assume that the two can be completely separated through a spatial filtering process and that the resulting error can be simply tolerated. Moderately complicated methods can at least estimate and attempt to compensate for the spatial-frequency cross-talk of the two effects, while ultimately, a full three dimensional model of the atmospheric phase distortion can be built which could form the basis for a Van Cittert type iteration between ray-traced scintillation and phase distortion (the derived function) and measured scintillation and phase distortion (based directly on the data). Again and as always, we have a variety of sophistication we can apply to a given problem, each trading off cost with accuracy. [0235] For present purposes we shall utilize a simple methodology, that of merely low-pass filtering both the reference image and each distorted image, then adding the difference image between the two onto the distorted image, producing a brightness-normalized distorted image with much of the scintillation effects removed (though some of the geometric distortion effects will be removed in the process). As such, we find that we have performed a gross separation of the two effects. The difference image, as a ratio to the reference image, is roughly a measure of the field-variant scintillation function for each given sub-aperture, though both cross-talkcd geometric distortion and imprecise low-pass filtering will introduce errors. The scintillation-normalized distorted image will be left largely with pure geometric distortion and will be input to the methodologies of the next sections, though cross-talked scintillation and imprecise low-pass filtering will also introduce some errors. [0236] Details on the exact steps involved in attempting to remove the scintillation function are contained in FIG. 31. Given a data image such as that in FIG. 30, step [0237] Obtaining a Geometric Distortion Estimate Between a Reference Undistorted Image and a Distorted Image: 2 Approaches [0238] The following two methods each produces an estimate of the geometrical distortion obtaining between two generally identical, other than geometric distortion, images. Having used the pre-processing method above of normalizing out the scintillation effects, we can now assume that the error sources are simple noise and blurring and not scintillation. The problem here addressed has generally been described in FIG. 29, where we will be searching for the bottom right diagram, given the upper left and the upper right as inputs. [0239] The Iterative Fat-To-Thin Line Scan Method [0240]FIG. 32 makes a basic point before diving into the main outline of this general approach. FIG. 32A is a simple one dimensional brightness distribution. FIG. 32B is the same distribution which has been geometrically distorted. FIG. 32C shows the integrals of each, running left to right, overlaid on top of each other. The primary point of this exercise is to note that the displacements of the ordinates between the two integral curves is a direct measure of the geometrical distortion between the two images—that is, in the case of pure non-inverting distortion. [0241] For one dimensional brightness distributions, therefore, it is a rather simple operation to measure the distortion: merely plot out the graph as in FIG. 32C), and then directly generate the distortion plot (FIG. 32D) based on the abscissa differences of each and every ordinate, plotted on the scale of the abscissas. [0242] Moving on to the two dimensional case, the situation becomes more complicated. It is only mildly more complicated if the typical distortions are on the scale of one pixel or less, and rather complicated if the distortions are larger than one pixel and the image has high contrast, such as the case in FIGS. 29 and 30 (assuming the pixel spacings are much finer than the box spacings). Fortunately, in atmospheric distortion conditions the distortion is generally not so pronounced as depicted in FIGS. 29 and 30, and indeed is typically on the order of one pixel in extent (depending on the specific optical design, however). Thus, this first described method of the fat-to-thin process is worthy of consideration. [0243] In a sense, the design of a high signal to noise ratio system prefers larger distortions because this will enable better precision measurements of that distortion. This is to say tnat the desire for minimal distortion in the service of simplifying the method of finding the correct distortion is in conflict with the desire for obtaining a higher signal to noise ratio on finding that distortion. [0244] When we apply a straightforward straight line integral to both images, the integral of the distribution along any given line on the distorted image has brightness values from both sides of the line adding spurious signal to the overall integration, thereby rendering the one-dimensional method of FIG. 32 useless when applied uncritically. The supposed geometric distortion being found can either be actual distortion or actual signal merely warped in from one side of the line or the other. [0245] The line integral method, however, can be salvaged if we approach the estimation of the two dimensional distortion in an iterated fashion, first finding gross and very low spatial frequency distortion, then refining the spatial frequency estimation as we integrate along “curved lines” which take into account the distortion profile on the orthogonal axes and which have been roughly estimated from previous iterations. Ultimately this approach is one which must be justified on an empirical basis for any given turbulence situation. [0246]FIG. 33 helps us take a pass through the required steps. The method begins (FIG. 33A) by taking broad line integrals along the orthogonal axes, where broad refers to weighted spatial averaging along the perpendicular axis to the direction of the line integral path. The resulting integral is further low-pass filtered and is compared to the same process performed on the reference image. This process is applied to a series of integrals first sweeping horizontal line integrals along the vertical axis, then sweeping vertical line integrals along the horizontal axis. The line integral shown in FIG. 33A is of the former variety. The total distortion found after a first full sweep is made is then assumed to be a rough first guess at the distortion. [0247] The spatial averaging weight function is then thinned up, and the new integration path is taken along the estimated distortion lines from the previous pass (FIG. 33B). In this way, a new distortion estimate is formed with slightly higher spatial frequency accuracy, while at the same time following a better estimate of the proper integration path. After a few or several iterations, the method should converge upon the true geometric distortion (FIG. 33C). As with all iteration routines, convergence can be regulated by appropriate damping variables, such as updating estimates by only one half their measured difference values, thus ensuring proper convergence. [0248] (In FIG. 33, the thickness of the lines represent the width of the Gaussian weighting functions along the orthogonal axis to the direction of travel of the line integral. The line integrals are first swept down the y-axis as shown, then across the x-axis, together sweeping across the entire two dimensional distortion. It may look simple shough, but we must remember that there will be noise, error, and residual scintillation effects in real data, not to mention the potential difficulties of slow or local minimum convergence.) [0249] The Global Search Method [0250] The second method of estimating geometric distortion, the global search method, is rather straightforward, but has the potential disadvantage of large computational loads and the difficulties associated with search algorithms including slow convergence times and the pitfalls of local minima. Nevertheless, search methods are also known to function properly if these difficulties are well considered, and hence the following technique is a viable one for determining the distortion function. [0251] Generally, the unknown variables can be any set of functions which adequately describe the two-axis shifting of an arbitrarily dense two dimensional array of control points, (i.e. reseau points). More simply, the unknowns can simply be the x-y offsets of some small grid of control points, with inter-spacing points appropriately interpolated, such as the bottom right drawing of FIG. 29. [0252] The search can have as the primary metric the mean squared difference between the “redistorted” image, i.e. the distorted image geometrically warped by the current search estimate vector, and the reference image. Pixels between the control points are, as mentioned, stretched via interpolated amounts, and inter-pixel interpolation values are generally useful for more accurate results. [0253] For the sake of speed, one pass of the “fat line integral” approach of the previous section can be used to get a grossly correct starting point for the search algorithm, and in the early stages only a certain percentage of pixel values, say one in every four, need to be used to calculate the mean squared error. [0254] Though this search method is a bit more complicated than given credit for in this summary, it is generally considered to be a mature prior art and the detailed implementation is a matter of typical refinement within that prior art. [0255] Notes on Anticipated Better Mode [0256] There is a third method of determining the geometrical distortion between two images that is presently being worked on but has not yet been reduced to practice and verified. It is anticipated that this method will eventually be the preferred embodiment. It goes as follows. [0257] The Line Integral Space Averaging Method [0258] The method described here is a fully analytic method, to use this term a bit loosely, in that it is a straightforward single iteration method of finding the precise geometric distortion function. Despite this advantage, it is strongly suggested that benchmark testing be performed on all three methods, or any newly devised ones, finding out which method has the least computational overhead and ensuring that they all generate substantially similar estimates of the geometric distortion. In order to fully describe the line integral space method, some background must first be discussed. [0259]FIG. 35 [0260]FIG. 35 [0261] (In more detail, FIG. 35C shows the definition of a Line Integral Transform of a given two dimensional function f. F is a function of r, theta, which are the usual polar coordinates, plus the r=0, as defined above. The integral is indefinite and is a line integral taken over the straight line defined by r, theta. Generally speaking, spatially bound, continuous functions, f, will have a unique line integral transform. The inverse transform is merely treating all points and values in F as an infinite system of lienar equations and solving for f. In spatially discrete digital images, however, a finite set of equations will suffice.) [0262] Where is this taking us? Well, the first thing we need to realize is that it is possible to transform back from the new space onto the conventional euclidean space. This can be done, to be overly pedantic, using a Hilbert-like transform where each point in the new space is explicitly the addition of the function values along the line in the euclidean space, thereby defining an infinite set of equations with an equally infinite set of unknowns, the classic uniquely mapped Hilbert formulation. Getting back to the simpler discrete world of digital images, it reduces to a large but nicely intuitive set of linear equations. For distortion problems in particular, where the distortions are generally of low spatial frequency relative to the pixel spacing, this set of equations can become quite small and tractable. [0263] We next define a new line integral called a radial line integral over the transform space. The “radial” refers to the fact that this line integral has a constant theta in transform space, and only these types of integrals will be utilized in the following steps. The integration is performed on F, not f, and is performed in the Line Integral Transform Space. Note that any given delta function in our original function first becomes a quasi-circle in LIT space (quasi because of the fat zero), and that any given radial line integral will be a step function with the step occurring at the appropriate point. [0264] We then take a set of radial integrals equally spaced in theta space. Each integral is performed both on the reference image and on the distorted image. The differences of the abscissas on any given ordinate is then interpreted as the integrated warp caused by the geometrical distortion, just as it was in FIG. 32. For any given radial integral, these differences are found and plotted out along the radial line itself. In this way, a new distribution is found plotted out in the line integral transform space. This is the line integral transform of the true geometrical distortion. It remains to set up this new distribution as a discrete set of equations and to solve this set of equations. [0265] The resultant radial integration differences are then fed into a set of linear equations in order to transform them back onto our normal euclidean space. Solving for this matrix solves for the distortion function. [0266] Some Final Notes on Inverting and Non-Inverting Distortion [0267] It should now be clear that the preceding methods make use of the assumption of non-inverting distortion. This is not to say that they will not function on inverting distortion, they will merely begin to exhibit different and generally rapidly increasing errors in the results that they produce. It will be important for engineering considerations to balance the needs of signal to noise ratios with the statistical onset of these error sources deriving from inverting distortion, and to do so as a function of the prevailing strength of turbulence; which, as already mentioned, is not much of a consideration for normal nighttime atmospheric turbulence conditions. [0268] Functioning on a Sparse Star Field or an Image with Grossly Non-Uniform Contrast [0269] A very important general case is now described, which also leads to further details on the implementation of the preceding methods. The salient features are drawn in FIG. 34. Essentially, the determination of geometric distortion is quite contingent upon either or both of A) local contrast, or B) anisoplanatism. The two actually reinforce each other. That is, if both are present, the signal to noise ratio of the measurement of the geometric distortion becomes that much better. As importantly, there must be sufficient detected photons in order to reach some nominal signal to noise ratio performance. [0270] Both methods described above must have modifications made when considering objects which are either sparse or which have grossly non-uniform contrast characteristics. Indeed, the modifications described herein should be at the core of the preferred embodiment when using any of these methods. [0271]FIG. 34A shows a figure which has two randomly distributed point sources and two extended sources as an example. (The upper left extended object has lower constrast than the object at the lower right.) Inspection of the image shows that only at the places where there is an optical signal can the measurement of geometric distortion take place. The implications of this are that only reseau points in the vicinity of the optical signal will get decent signal to noise ratios on their shift measurements. [0272] There are two basic ways of dealing with this: either use all the reseau points as if an entire S/N-ratio-homogenous image is present, or respace the reseau points to “bunch around” the areas of maximum signal, so that all reseau points have generally equal signal to noise ratios. Ultimately, both approaches will get to the same end point. [0273] The point about the signal to noise ratio being contingent on either or both of Al local contrast, and B) anisoplanatism is somewhat arcane but relatively important. Conventional wisdom has it that zero-contrast extended objects do not give rise to a shift signal in wavefront sensors, whereas FIG. 34, as well as the detailed discussion surrounding FIG. 26A and 26B of the earlier disclosure, has shown that anisoplanatism itself creates a shift signal even on zero-contrast objects with at least some brightness. This fact alone is a performance differentiator between prior art mono-axial wavefront sensors and the wide field wavefront sensor of this section and the full field version of a previous section. [0274] Turning to a detailed look at FIG. 34B, the top line integral has a waveform of relatively constant positive slope, where it is labelled as ‘Anisoplanatic signal only’. This means that only the squeezing and stretching of the signal due to anisoplanatism will create a shift signal in this region. On the other hand, the lower line integral has a ‘bumpy’ signal in the extended object, so that both the squeezing and stretching of the signal due to anisoplanatism and the movement of the object features will create a shift signal in this region. Thus, the signal to noise ratio on the measurement of the geometrical distortion will be higher on the high contrast extended object than on the low contrast extended object. [0275] (FIG. 34B depict two line integrals showing how point sources and the higher contrast extended object provide for high “shift signal” of geometric distortion, where the low contrast object still has “shift signal” but must rely purely on anisoplanatism to create the signal.) [0276] Finally, the scintillation function measurement is almost exclusively related to just the primary local photon count and has little to do with the local contrast. Thus, at the end of the day, the spatial non-uniformities of the signal to noise ratios on the scintillation function and on the geometric distortion functions are only slightly correlated. This fact, however, is of little theoretical concern to ultimate system performance. [0277] Optimizing the Optical Design for the Wide Field Hartmann-Shack Wavefront Camera [0278] Generally speaking, it is important to vary the parameters of the optical elements in order to maximize signal to noise ratios and minimize spatial filtering on pupil function estimates, all while minimizing the number of photons required. FIG. 8 had a complete example specification for the full field wavefront sensor. Here we explore a typical example of specifications for the wide field Hartmann-Shack camera. [0279] The good news is that for most astronomical observations from nominal sites, inverting geometric distortion is quite rare. Empirical proof can be readily obtained by merely imaging a binary star system and observing that the anisoplanatism of the tilt—along the radial arm connecting the two images—is quite small and that the two stellar images effectively vibrate in unison. On very rare occasions, for our hypothetical site, the two images will merge or cross. The problem of inversion will become a significant issue only when the ambient seeing conditions become quite poor, such as r [0280] On to the more practical manters, a fundamental consideration in the design of the optical elements of the wide field Hartmann-Shack camera is trading off the signal to noise ratio on determining local geometric distortion sensitivity. In general, the precise measurement of the geometric distortion would like to see finer pixel spacings, while the desire to collect photons into a single photosite for the purpose of reaching dimmer beacons would dictate more coarse pixels “fine” and “coarse” refer to higher and lower magnifications respectively. [0281] The “signal amplitude” of typical geometric distortion in r [0282] Taking these very crude postulations, and positing once again the basic telescope arrangement of FIGS. 4 and 8, we can determine that a pixel spacing of 1 arc-second per pixel is reasonable for at least imaging brighter objects such as planets. Here we find roughly a five to twenty-five signal to noise ratio on geometric distortion signal to noise. On planets in particular, we must remember that we will have thousands of pixels giving us signal values, and the aggregate will thus give us even better signal to noise performance due to a kind of averaging process. We can also target a 36 by 36 arc second square field of view for both the primary detector and the wide field Hartmann-Shack camera, only because this will work out nicely for a 1024 by 1024 array on our 2.5 meter aperture telescope (Both Jupiter and Saturn can get a bit bigger than this, and for those cases, the field of view would need to be enlarged using the basic principles herefollowing. [0283] The basic optical elements are as they were in FIG. 4, except that the primary array [0284] A final note is in order concerning chromatic aberration of the lenslet system. In the previous parts of this disclosure it was seen that a relatively wide spectral band light beam is split into the wavefront camera. This is still the case with the currently described wide field Hartmann-Shack (WFHS) camera. Now, however, the WFHS camera is explicitly setting up geometrically distorted images of arbitrarily extended objects rather than centroid shifts on point or highly localized sources. The chromatic aberrations will not change between the two systems, but the underlying processing methodologies may at some point begin to experience a loss of signal amplitude due to this chromatic aberration. This effect should be noted in determining the maximum allowed spectral bandpass of the dichroic splitting arrangement [0285] Converting the Distortion Measurements into Field Angle Dependent Pupil Function Estimates [0286] The preceding processing methods and their associated optical design layout find both the scintillation function and the geometrical distortion function of the sub-aperture images, functions themselves of field angle. It remains to convert these measurements into the form required by the processing methodology depicted in FIG. 10. What we actually require is to turn these measurements into the point spread function estimates at the nominal field angle sample points depicted in FIG. 12A. [0287] In general, this is exceedingly simple. The distortion function thus found is a direct measure of the local tip-tilt, just as if a star was located precisely at each field angle sample point [0288] Now that we have a method for generating the field angle dependent pupil functions, we can replace the entire wide field Hartmann-Shack wavefront camera for the full field wavefront camera disclosed earlier and pick up where that disclosure left off. The full field camera may not be entirely obsolete, since some turbulence situations may be such that the optics cannot reasonably produce good statistical non-inversion, and the full field beam splitting design may offer some advantages in terms of simplifying and segregating the local potentially inverted distortions. This may also be of assistance in the quest for using dimmer and dimmer object beacons, where “allowing” inversion may increase sensitivity. Be this as it may, the simplicity and minimized optical surfaces of the wide field Hartmann-Shack arrangement should present a superior choice for most applications. [0289] Another Note on Anticipated Better Modes: Photon Counting Detectors and Cameras [0290] The lower noise a detection system has, the better. The choice of specifying CCDs for both the primary detector [0291] It should be recognized that the particulars of the detector(s) employed in an embodiment of the present invention are entirely maners of design choice for the system implementer; the present invention can be implemented with a number of known detector technologies other than those described in the foregoing paragraph [0292] Derivative Versions of Resolution Enhancement of Turbulence Degraded Images, for Use on Single Detector Telescopic Systems [0293] So far, this disclosure has described an imaging system having two detectors, one taking primary exposures and one taking wavefront distortion exposures. It so happens that a portion of the disclosed methodologies can “trickle down” to systems which only employ a single primary detector array. In practical terms, the vast majority of imaging systems have only a single detector (obviously), and thus a reasonable improvement in resolving power on such systems, in the presence of turbulence, might be of some use. In essence, the following method uses the same exposure data both as primary data and as wavefront data. By separating out the two functions of the data, a methodology is presented which should provide for reasonable resolution enhancement. Among many mitigating factors, it will be seen that the methods are best employed where an object scene does not change appreciably over, for example, a set of ten exposures. The reasons for this are that the methods presented here work best when a multiple set of image data sets are present. [0294]FIG. 36 gets us right to the heart of the methodology. FIG. 36A depicts a non-distorted image of an object; FIG. 36B depicts three examples of the same object with both geometrical distortion and spatially random blurring. (Note that the blurring randomly roams around the image, both across spatial features and image to image.) The images on the right are typical of, for example, short exposures (e.g. 10-100 milliseconds) on a 30 cm to 1 m aperture telescope at a typical mountain site. It also might be typical of a terrestrial imaging system, of a somewhat smaller diameter aperture and in somewhat higher turbulence conditions. The present section of the disclosure describes a method whereby a set of such images as on the right are taken and synthesized into a single clear image of the object. The physical setup is therefore like that of FIGS. 2 and 4, only now there is no dichroic splitter [0295] One novel aspect of the single detector arrangement disclosed herein is perhaps most clearly highlighted by comparison to certain prior art systems. In particular, there are a wide variety of prior art imaging systems which remove the global tip-tilt in a stream of video images. Specifically, there are cameras referred to as “steady-cams” which are well known to remove “motion jitters.” Likewise, cameras with either mechanical and/or image processing technologies have been made which can produce a steady image scene even when the camera is mounted on moving vehicles or aircraft. Finally, there are well known applications where a series of “shifted images” are gathered and registered with each other to effectively remove the “shift” between them, and then averaged together to form a single clear still image. The latter is typical of planetary imaging. In all of these prior arts, it is fair to say that “first order” global tip-tilt correction is being applied. A distinction with the arrangement described in this section is that the present arrangement corrects beyond this first order. [0296] In summary, the process described in this section performs a global registration of the distorted images, determines the local variations of tip-tilt in order to be removed (the geometric distortion), and then preferentially adds the areas of highest contrast via weighting criteria. The weighting function is meant to keep track of local contrast variations from one image to the next at a given location of the image, not from one part of the image to another. This acknowledges that often a given image can be crisp in one region and blurred in another all in a single exposure, and that the crisp regions and blurred regions randomly wander around the image plane from one distorted image to the next over a large series of E exposures. The point is that any given region will have its fair share of crisp data given enough exposures. [0297] In speckle imaging processing, “shift-and-add” techniques have been used in certain applications, such as examining close double stars; a simple centroid is determined, or a highest pixel value, the resulting x-y offsets applied and the images added. In hindsight, the arrangement of the present section may be viewed as a rather involved extension on this core idea, applied to arbitrary extended objects. Likewise, it may seem from the literature that shifting and adding slightly offset images may be used to reduce basic tracking errors and global tip-tilt in multiple exposure telescopic imaging. However, no specific description of such an approach is presently known to the inventor, and this single detector arrangement basically uses the global registration merely as a kind of pre-processing of the images, or, to use analytic vernacular, the simple removing of the lowest order of geometric warp—global tip—tilt. It is generally acknowledged by those practiced in the art that these simple shift-and-add techniques have their limitations and will only produce a limited resolution enhancement; thus the dual detector set-up of the earlier disclosure is believed necessary to achieve near diffraction limited performance on larger aperture telescopes and in more turbulent conditions. [0298]FIG. 37 steps through the single detector enhancement methodology. We begin once again by gathering a series of short exposure digital image data sets of the object, step [0299] Once two given images are “registered,” it remains to resample one image onto the grid spacing of the other. This too is a familiar image processing technique and reference is made to standard texts dealing with these basics. The images can then be effectively averaged together. The registered images now become the working exposures for the next steps. [0300] Step [0301] Step [0302] Step [0303] A few more ancillary notes are in order. First, short exposures are generally advised so long as they don't trade off other performance goals such as overall noise. The short exposures tend to “freeze” both atmospheric distortion and minor tracking error. If either of these effects last too long and are integrated into one given exposure, they tend to irreversibly destroy high frequency information in the raw exposures. Second, narrow band filtering can also be effectively employed. The filtering merely minimizes chromatic blurring on monochromatic detectors, again assisting in retaining high frequency information in the raw exposures. [0304] Conclusion [0305] This disclosure has aptly focussed upon an “optical telescope” operating in the open atmosphere as the preferred embodiment of the invention, and has numerous references to alternative embodiments. “Optical” has both explicitly and implicitly been used in its most general sense as referring to the entire electromagnetic spectrum. The wavelength of 500 nanometers was repeatedly used as an example central wavelength about which a narrow band of light is imaged. To use this wavelength as an example one final time, it can be seen that whereas prior art systems such as adaptive optics systems and speckle imaging systems could indeed achieve near diffraction limited performance at this wavelength and others like it, they could do so only within a steradian field coverage of 5 arc seconds in atmospheric conditions in the r [0306] It should once again be noted that, while the foregoing disclosure has addressed in considerable detail certain telescope applications, the teachings thereof are more generally applicable to any imaging system degraded by temporally or spatially varying distortion. [0307] The issue of alternative embodiments to those detailed herein has been quite a recurring theme in this disclosure. Those skilled in the art of telescope or imaging system design will be able to clearly recognize the range of design options available for creating other embodiments. Though this principle is almost a cliche in the spirit of patent protection, it is nevertheless quite important to recognize this freshly for inventions which must perforce borrow upon not only decades but centuries of prior art; inventions which in some embodiments are best described as very large systems. It was with this in mind that FIGS. 1, 5, and Referenced by
Classifications
Legal Events
Rotate |