US 20020150299 A1 Abstract A method for adaptive spectral sensing developed for a two-dimensional image made up of picture elements. The method calls for illuminating at least one of the picture elements with an input light and deriving a time-varying spectral signal from the input light for that picture element. The time-varying spectral signal is processed with a time-varying reference signal by using a mathematical function such as convolution, multiplication, averaging, integrating, forming an inner product, matched filtering, addition, subtraction or division to obtain a processed output value for the picture element and this output value is then used in determining a spectral characteristic of the input light. The time-varying spectral signal is conveniently derived by optical filtering of the input light yielding an optical time-varying spectral signal. This method can be used by itself or in combination with at least one other adaptive technique such as adaptive spatial sensing and/or adaptive temporal sensing.
Claims(23) 1. A method of adaptive spectral sensing for a two-dimensional image comprising picture elements, said method comprising:
a) illuminating at least one of said picture elements with an input light; b) deriving a time-varying spectral signal from said input light for said picture element; c) processing said time-varying spectral signal with a time-varying reference signal to obtain a processed output value for said picture element; d) employing said processed output value for determining a spectral characteristic of said input light. 2. The method of 3. The method of 4. The method of 5. The method of 6. The method of 7. The method of 8. The method of 9. The method of 10. The method of 11. The method of 12. The method of 13. The method of 14. The method of 15. The method of 16. The method of 17. A method of adaptive spatial sensing for a two-dimensional image comprising picture elements, said method comprising:
a) illuminating at least two of said picture elements with an input light; b) scanning a projection of said at least two picture element across a spectral detector element to obtain scan response signals from said at least two picture elements; c) processing said scan response signals with a reference signal to obtain a processed output value for said at least two picture elements; d) employing said processed output value for determining a spatial characteristic of said input light. 18. The method of 19. The method of 20. The method of 21. The method of 22. A method of adaptive temporal sensing for a two-dimensional image comprising picture elements, said method comprising:
a) illuminating at least one of said picture elements with an input light; b) deriving a time-varying spectral signal from said input light for said picture element at two different points in time and averaging; c) processing said time-varying spectral signal with a time-varying reference signal to obtain a processed output value for said picture elements; d) employing said output value for determining a temporal characteristic of said input light. 23. The method of Description [0001] This application is based on provisional patent application No. 60/226,046 filed on Aug. 16, 2000 which is hereby incorporated by reference. [0002] The present invention relates generally to adaptive spectral, spatial and temporal sensing in imaging applications and is especially useful in two-dimensional imaging arrays. [0003] Conventional cameras and photographic systems allow limited colors in images. Color cameras typically collect information in broad spectral bands assigned to each color, and these spectral bands are fixed. There are also imaging systems which operate in spectral bands outside of the visible range, such as in the infrared spectral region. Such imaging systems are very useful in that they enable thermal imaging using the emitted infrared radiation from warm bodies, for example, and specific spectral bands in the infrared also allow identification of chemicals and materials. More specific identification is possible if narrow spectral bands are used, or if specific collections of spectral bands are used that are well chosen to discriminate between different chemicals or materials. Such discrimination can be very useful in identifying targets, tracking biological and chemical agents, finding materials, locating plants and in tracking various other materials and objects in the environment. [0004] Hyperspectral imaging attempts to collect image data at a very large number of colors or in many different spectral bands. This technique generates very large amounts of data that is difficult to process. An alternative approach is to make a system that collects a few specific wavelengths using fixed filters in front of detector elements. Such a system can detect specific kinds of objects with specific emission, reflection or absorption spectra. However, this type of system is typically only usable for the specific objects it is designed to detect. [0005] Another approach uses a kind of adaptive spectrometer, as discussed by Deverse at al., “Spectrometry and Imaging Using a Digital Micromirror Array”, American Laboratory, Vol. 30, No. 21, 1998, pp. S112. This approach allows the optimum spectral response to be chosen to discriminate objects. This system is based on the use of micromirror arrays and can provide choice of spectral sensitivity at least for a linear array of light spots or picture elements. Unfortunately, it is more difficult to use this technique for controlled spectral sensitivity in a two-dimensional image. [0006] What is required is adaptive spectral sensing which is easy to implement for two-dimensional arrays of picture elements. It would also an advance to provide for additional adaptive sensing capabilities, such as adaptive spatial sensing and adaptive temporal sensing. [0007] Accordingly, it is a primary object of the present invention to provide for efficient adaptive spectral sensing in two-dimensional arrays of picture elements. [0008] It is another object to provide for efficient adaptive spatial and temporal sensing in two-dimensional arrays of picture elements. [0009] These and other objects and advantages will become apparent upon reading the ensuing description. [0010] The objects and advantages set forth are achieved by a method for adaptive spectral sensing developed for a two-dimensional image made up of picture elements. The method calls for illuminating at least one of the picture elements with an input light, e.g., light from an object to be examined, and deriving a time-varying spectral signal from the input light for that picture element. Next, the time-varying spectral signal is processed with a time-varying reference signal to obtain a processed output value for the picture element. The output value is then employed for determining a spectral characteristic of the input light. The spectral and reference signals are processed by applying to them a mathematical operation such as convolution, multiplication, averaging, integrating, forming an inner product, matched filtering, addition, subtraction and division. [0011] The time-varying spectral signal is conveniently derived by optical filtering of the input light yielding an optical time-varying spectral signal. The optical filtering can be performed by an optical filter positioned in front of the picture element. In one embodiment, the optical filter is a scanning optical filter and the optical filtering function is performed by scanning. In another embodiment, the optical filter is a Fourier transform scanning optical filter and the optical filtering function involves performing a Fourier transform. The optical filtering can also be combined with other operations, e.g., optically splitting the input light. [0012] Alternatively, the time-varying spectral signal is derived from electrically controlling a spectral detector element. [0013] The spectral detector element can be any suitable photodetector such as a quantum well infrared photodetector, a silicon photodetector or an analog electronic multiplier. In the process of electrically controlling the photodetector the derived time-varying spectral signal can be an electrical time-varying spectral signal such as a voltage, a current, an inductance, a power, an electric field, a magnetic field, a resistance, a capacitance or an inductance. The time-varying reference signal with which the electrical time-varying spectral signal is processed is also in the electrical form; i.e., it is an electrical time-varying reference signal. [0014] In a preferred embodiment, the two-dimensional image of picture elements is formed of an array of pixels. In other words, the picture elements of the image correspond to a number of pixels. It is also convenient that in this embodiment time-varying spectral signals be derived for each of the pixels. The time-varying reference signals used in this derivation can be different for different pixels. [0015] The technique for adaptive spectral sensing can be implemented independently or together with at least one other adaptive sensing technique such as adaptive spatial sensing and adaptive temporal sensing. [0016] The specific embodiments of the invention are described in the detailed description with reference to the attached drawing figures. [0017]FIG. 1 is an exploded isometric view illustrating the general principles of an adaptive spectral sensing system according to the invention. [0018]FIG. 2A is a schematic diagram of a portion of an alternative adaptive spectral sensing system. [0019]FIG. 2B is a schematic diagram of a portion of still another alternative adaptive spectral sensing system. [0020]FIG. 3 illustrates a plan side view of another adaptive spectral sensing system. [0021]FIG. 4 is a circuit diagram showing a photoconductive detector of the system shown in FIG. 3. [0022]FIG. 5 is a timing diagram illustrating an example of an optical time-varying spectral optical signal from the scanning optical filter from an object of interest, and a corresponding time-varying reference signal. [0023]FIG. 6 is a diagram of a Fourier transform scanning optical filter. [0024]FIG. 7 is a graph of responsivity (photocurrent per unit optical power) of a quantum well infrared photodetector at a temperature of 77 K as a function of bias voltage. [0025]FIG. 8 is a circuit diagram of a simple multiplier circuit. [0026]FIG. 9 is a diagram illustrating the imaging of an object for adaptive spatial processing. [0027]FIG. 10 shows a simplified schematic of the use of Risley prisms in an imaging system. [0028]FIG. 11 is a diagram illustrating the locus of beam positions when rotating two Risley prisms of equal wedge angle, with one prism rotating 5 times as fast as the other. The units of displacement are the magnitude of the displacement the beam would experience in passing through one such prism. [0029]FIG. 12 is a graph of a reference function as a function of position traced out as a function of time t as two Risley prisms are rotated at rotation rates of f and 5f respectively. Both prisms have equal wedges. The value of the reference function is positive in the center and negative at the sides. The vertical axis is the amplitude of the reference function. [0030]FIG. 13 is a diagram illustrating the construction of a total combined reference function by multiplication of three separate reference functions for the spectral, spatial and temporal properties desired. [0031]FIG. 14 is a circuit diagram of an exemplary time averaging circuit. [0032] The principles and operation of the invention will be best understood by first examining an adaptive spectral sensing system [0033] Preferably, picture elements [0034] System [0035] A control unit [0036] Optical filter [0037] There are various ways in which modulator [0038] The operation of system [0039] Time-varying spectral signal [0040] Time-varying processed signal [0041] Control unit [0042] The output value for pixel [0043] The manner in which system [0044] Control circuit [0045] A modulator [0046] System [0047]FIG. 2B illustrates a portion of another alternative system [0048] System [0049] In any of the above systems it is understood that the spectral sensitivity could be set the same for every picture element or pixel if desired. This would be done by feeding the same time-varying reference signal for each pixel. Alternatively, the spectral sensitivity could be set to differ from pixel to pixel, if desired, by feeding different time-varying reference signals for different pixels. It is also understood that the same pixel can have several different spectral filtering operations performed on it to generate multiple different spectral sensitivities, just as a conventional color camera returns values for each of three different colors in each pixel. [0050] Several different spectral filtering operations can be achieved, for example, by optically splitting the input light before the scanning filter onto multiple separate systems, each configured for one kind of spectral response. Alternatively, the optical time-varying spectral signals can be optically split after the scanning optical filter into multiple copies of the optical time-varying spectral signals. The copies are then processed separately with different time-varying reference signals. Appropriate beam splitting techniques to split the input light or obtain multiple copies of the optical time-varying spectral signal are well-known in the art. When operating in the electrical domain, the electrical time-varying spectral signals can be processed with different electrical time-varying reference signals and separately time averaging or integrating each result to generate multiple different electrical processed output values. Yet another solution available in the electrical domain is to operated with multiple different electrical time-varying reference signals one after the other in sequence to generate a sequence of different electrical processed output values, each corresponding to a different spectral filtering of the corresponding picture element or pixel. [0051] It will also be clear to a person skilled in the art based on the above-described embodiments, that various functions such as multiplication and time-averaging or multiplication and conversion between optical and electrical signals can be combined in one system or circuit if that is more convenient than separating these functions. It is also possible to perform some or all of the optical filtering and processing functions by one device on all pixels at once. In system [0052]FIG. 3 illustrates an adaptive spectral sensing system [0053] System [0054] During operation, an input light [0055]FIG. 4 illustrates the electrical connections for an exemplary photoconductive detector [0056] It should also be noted that photoconductive detectors can be made with many different kinds of materials, including semiconductors such as silicon, gallium arsenide, indium gallium arsenide, indium phosphate, lead sulfide, lead selenide, lead tin telluride, indium antimonide, mercury cadmium telluride, and in general with most known semiconductor materials, as is well known to those of ordinary skill in the art. The specific material choices will depend on the portion of the spectrum of interest in input light [0057]FIG. 5 is a simple example of an optical time-varying spectral signal from the scanning optical filter from an object of interest, and a corresponding time-varying reference signal. When the scanning optical filter is a simple spectral filter such as a Fabry-Perot filter, the signal shown would correspond to the optical spectrum from the object of interest consisting of two spectral bands. The corresponding reference signal in this example is of similar form to the example optical time-varying spectral signal, except that it is somewhat negative in spectral regions outside the spectral bands of interest (see zero point). In this example, this negative value is chosen so that, when a uniform spectrum is presented to the input of the system (instead of the optical signal form the objects of interest), the time-integrated or time-averaged result of the multiplication by the time-varying reference signal is zero. Hence, in this example, the desired signal is detected and a uniform spectral signal (e.g., background noise) is rejected. Other forms of time-varying reference signal can also be used to reject other forms of undesired input spectrum while preserving sensitivity to the object of interest. [0058] One simple mode of operation of the system is to scan the scanning optical filter repetitively. In such a mode, the desired processed output value can be obtained by simple time-averaging of the multiplied time-varying signal. Note that it is not necessary that the rate of scanning be constant, or even that it be unidirectional; appropriate choice of the reference signal can compensate for any such scanning function. [0059] One simple method of constructing a time-varying reference signal to correspond to a particular spectrum produced by from an object of interest is to use such a source as the input to the system and to record the optical time-varying spectral signal as a function of time. This can be done, for example, by detecting the optical signal to construct an electrical signal and recording the electrical signal as the scanning optical filter is scanned through a complete cycle. The resulting time-varying signal can then have a constant added to it so that it has zero net integrated area. This signal can then be used as a time-varying reference signal that detects the spectrum of interest while suppressing a uniform spectral background (noise) Such a method also automatically compensates for finite temporal response in the detector itself. It will be understood by a person skilled in the art that other methods of constructing time-varying references signals can be more appropriate, depending on the application, system and other parameters. [0060] It should also be understood that it is not necessary that the multiplication by the reference signal correspond exactly to a standard arithmetic multiplication. The multiplication process can be nonlinear, with the result depending in a nonlinear fashion on the amplitude of the time-varying reference signal. Such nonlinearity can be compensated by an appropriate change in the time-varying reference signal. [0061] An alternative approach to a scanning optical filter is to use the method known as Fourier transform spectroscopy. One simple version of such a Fourier transform scanning optical filter [0062] For any of the systems described herein the preferred mode of operation is at an output rate corresponding to the frame rate of a typical imaging system, which could be in the range of 30 frames per second. Each frame time ({fraction (1/30)} of a second in this illustrative example) would therefore preferentially correspond to one or more cycles of the scanning optical filter so that the signal for each pixel in the image would be a sufficiently averaged result of the process of spectral filtering and multiplying by the reference signal. Such a cycle time for a scanning filter is readily obtained by vibrating bodies, for example. Operation at cycle times much faster than {fraction (1/30)} of a second is also readily feasible with many vibrating devices. One example of a device capable of vibrating at such frequencies is a loudspeaker cone. Interferometers similar to the Fourier transform scanning optical filter [0063] Methods for generating a time-varying reference signal are well known to those skilled in the art. Such a signal can for example, be stored as a set of numbers in a digital processor, and converted to an analog signal (such as an electrical voltage) by sequentially reading out the numbers into an analog-to-digital converter. [0064] It is not necessary that scanning spectral filter scan the spectrum in a linear fashion or measure each spectral region with equal accuracy. For some applications, such as a situation where the spectral information of interest is concentrated in a few small spectral areas, it would be preferable to have the filter scan more slowly through the spectral region of greater interest. Hence the spectral scanning itself can be adapted to suit the problem of interest, as well as choosing the time-varying reference signal appropriately. A person skilled in the art will recognize that it can also be useful to have a Fabry-Perot filter of adjustable finesse, for focusing in on these regions of interest. [0065] One convenient choice of photodetector for operation in the infrared wavelength range is the quantum well infrared photodetector (QWIP). QWIPs are photoconductive detectors, and hence are well suited for use with the system of invention. For example, in the region between 0.5 Volt and 1.5 Volt bias, the photocurrent for a given optical power increases smoothly in the example results indicated in the graph of FIG. 7. These results are further described in the article “Semiconductor heterostructures: a quantum Lego for the infrared,” by E. Rosenher in [0066] Silicon is useful as a photodetector for use in systems of the present invention from the near infrared to the ultraviolet. Its use as an ultraviolet detector is discussed by M. Razeghi and A. Rogalski, “Semiconductor Ultraviolet Detectors,” J. Appl. Phys. 79, 7433-7473 (1996). Ultraviolet detectors can be particularly effective for identifying biological materials. Silicon is extensively used in CCD (charge-coupled devices) cameras in the visible range, and silicon detectors are also used in the more recent CMOS cameras that use processing compatible with silicon CMOS integrated circuits (see, e.g., M. J. Loinaz, K. J. Singh, A. J. Blanksby, D. A. Inglis, K. Azadet, and B. D. Ackland, “A 200-mW, 3.3-V, CMOS Color Camera IC Producing 352×288 24-b Video at 30 Frame/s,” IEEE J. Solid-State Circuits, 33, 2092-2103 (1998)). Such photodetectors can be used with circuits well known to those skilled in the art to produce currents or voltages in response to input light. [0067] It is also possible to make silicon photodetectors whose output signal varies in response to an electrical control signal. One way to do this is to vary the depletion length, and hence vary the photocurrent collection efficiency, hence allowing a multiplication of the input signal by a reference signal. It is possible to make a silicon photoconductor by using a field-effect transistor channel as the photoconductor, with one end of the channel connected to a low impedance preamplifier (to amplify the detected current signal), and the other end to the reference signal. The number of carriers in the channel would depend on the number of carriers (electrons and/or holes) generated by optical absorption at or near the channel region. The electrical current signal collected would tend to be proportional to the number of such carriers and to the reference voltage signal. [0068] There are many circuits capable of multiplying two electrical signals. In the present invention, the two signals of interest are the time-varying reference signal and a voltage or current signal output from a photodetector corresponding to the time-varying spectral signal. One very simple form of multiplier circuit [0069] Any system of the invention can be adapted for sequential processing of different spectral signals. By changing the reference signal in successive periods of the scan of the spectral filter, successive different output values will correspond to different spectral filterings, allowing multiple such filterings to be performed by the one detector element. [0070] The same basic kind of approach taken above for adaptive spectral sensing and processing is also applied to adaptive sensing and processing in the spatial domain, and in the time domain. [0071] The underlying concept here is to scan the image over the detectors, weighting different parts of the scan differently in averaging the resulting output from the detectors. This implements a convolution of the image with a kernel, a basic and important class of image processing applications. The method is illustrated in FIG. 9. This figure illustrates an object [0072] For example, consider a particular detector element that, in the absence of rotation of mirror [0073] If mirror [0074] So far the kernels and scanning operations were discussed as if they correspond to discrete steps between specific pixels, but it should also be understood that they can also be represented as continuous functions, and the scanning can be continuous rather than step-wise. Such a transition to a continuous domain does not affect the basic concept of convolution with a kernel. [0075] Note that changing the scanning range of the mirrors effectively “dilates” the kernel; that is, one can choose to have the kernel operate very locally, with a small scanning range, or over a larger fraction of the image by using a larger scanning range. Note too that it is not necessary that the scanning be linear or at a constant rate in time. Different scanning patterns can be compensated by redesigning the reference function to compensate for the different patterns. Such scanning at non-constant rates can be advantageous. For example, in the Laplacian kernel described above, the scan can be designed to spend 8 times longer in the central pixel position than in each of the others, in which case the signal to noise ratio of the whole process could be improved because 8 times as many photons would be detected on the average in this central pixel position, in correspondence with the magnitude of the weight being applied to this position. [0076] Various other image processing kernels can be implemented this way. Many of these are discussed in the book “The Image Processing Handbook” by John C. Russ (CRC, Boca Rotan, 1992). Examples include a sharpening operator with a kernel of the form:
[0077] intended to sharpen the edges in an image while still leaving the image itself visible, and a smoothing operator of the form:
[0078] that averages out local variations (versions of this kind of operator could be used for averaging out noise, or for averaging out local texture, (e.g. foliage)). Simple spatial derivative operators in one dimension are also straightforward to implement, for example with the one-dimensional kernels:
[0079] or versions in 3×3 matrix form such as:
[0080] Note that the Laplacian operator essentially represents a high-pass filter for spatial frequencies, and the smoothing operator essentially represents a low-pass filter. There are other similar operators that will perform similar functions. For example, the difference-of-Gaussian operator, which takes the difference between two different Gaussian smoothing operators of different size, also performs a high-pass filtering, and has a form similar to the Laplacian operator. Many different kernels exist that also perform local smoothing. [0081] A yet more sophisticated concept is to use multiple different kernels that may even constitute a complete set. For example, in one direction we could apply the following set of kernels, one after the other:
[0082] These constitute a complete set of kernels (any 4 element kernel (or vector) can be made as a linear combination of the kernels), and can be viewed as a simple discretized Fourier transform basis (related also to a Hadamard transform basis). We can then view any particular spatial frequency filtering in terms of a weighting of the result from multiplying by these 4 basis kernels as reference functions. Of course, alternatively, we could choose that weighting in advance to construct that one kernel, though then we would lose the ability to perform subsequent filtering operations on the sequence of outputs. Various other complete sets are possible. This concept has been described for a particular case of two-dimensional kernels by, for example, W. Frei, and C. C. Chen, “Fast Boundary Detection: A generalization and a new algorithm”, IEEE Trans. Comput. C-26, 988-998 (1977). [0083] It would also be possible to construct more complex kernels designed to look for a specific shape in the image. [0084] Use of a micromechanical array of mirrors or other devices for performing the spatial scanning separately for each pixel would permit spatial adaptation of the whole array, with different spatial scanning for different pixels. The same would be true for spectral filtering if different scanning filters were constructed for each pixel, allowing possible high level adaptation for singling out particular kinds of areas. [0085] Note that this approach avoids conversion to a coherent signal for Fourier domain optical image processing by instead applying the same kernel to every point in the original (possible incoherent) intensity image. [0086] Risley prisms are wedges of transparent material, usually with some small angle between the two faces. A beam passing approximately perpendicularly through one of such wedges has its angle changed by a fixed amount, with the direction of the angular change depending on the rotational orientation of the wedge about an axis approximately perpendicular to the faces of the wedges. If two such prisms are put one after the other in the light beam, by changing the rotational angles of the prisms, it is possible to manipulate the angle of the beam continuously in two directions over the same range. Hence, such two so called Risley prisms can be used as an alternate method to scan the angle of a beam. They have the advantage that they can be placed “in-line”, not requiring a large angle change as required by the simple mirror of FIG. 9. [0087]FIG. 10 shows a simplified schematic of the use of Risley prisms [0088] With such a pattern as in FIG. 11, it is possible to construct a simple reference function that will result in the approximate evaluation of a Laplacian-like operator. Choose a function −cos 8πft (where t is time) is chosen as the reference function (i.e. a cosine at frequency 4f) where f is the rotation rate of one prism, and 5f is the rotation rate of the other prism. Then the value of the reference function as a function of the position in the two spatial directions is as shown in FIG. 12. [0089] This reference function therefore weights the center of scan range strongly positive, and the wings negative, with an average value of zero. Note that the center point is accessed 4 times altogether, whereas each extreme edge point is accessed only once in the complete cycle of the system. A cycle corresponds to a time 1/f. We can therefore view this operation as being approximately equivalent to operating with the kernel
[0090] which is a simple version of a Laplacian operator. [0091] One can also take temporal derivatives by appropriate combinations of successive measurements. For example, to find the first derivative in time (the rate of change of the pixel in the image) we can multiply by the same function in two successive periods, but with opposite signs. If there is no change in the image in these two successive periods, the net average is zero. Only if there is a change is the average different from zero. Hence, we need only change the time-varying reference function to have this two-period inverted form to achieve temporal differentiation of the image. [0092] One power of this approach to spectral and temporal processing of image information is that all of these different aspects can be combined. Thus, for example, it is possible to look for objects with particular spectral features where the object has sharp edges and is moving. The key to this is to construct appropriate reference functions. [0093] The reference functions are particularly easy to construct if the different modalities (spectral scanning, spatial scanning, and temporal scanning) are run at very different scan rates that are integer multiples of one another. This is illustrated in the example of FIG. 13. In this example, the spectral scanning is performed at a relatively high scanning rate, with 6 successive spectral scans performed in the time shown in FIG. 13. Next fastest is the spatial scanning function, which in this case is chosen as a simple one-dimensional second spatial derivative (Laplacian); here the physical scan is presumed to be linear and unidirectional in time in one direction (though that is not in general a necessary set of restrictions), so the reference function is chosen simply as the sequence of values −1, +2, −1. The slowest of the three reference functions is here chosen as the temporal derivative function, which has successive values +1 and −1, though each value persists for three periods of the spectral scan. The reference function for the spectral scan here is intended to have an average value of zero, as do the 1-D spatial Laplacian and temporal derivative reference functions. The final total combined reference function is the product of the three and also has an average value of zero. In this example, we choose to integrate over six periods of the spectral scan, so each function is integrated for a complete number of periods. [0094] This example total reference function, when multiplied by the signal and integrated, will therefore tend to find pixels with a particular spectral signature, near an edge (so there is a strong Laplacian), and changing in time, giving large integrated signals in this case. Constant pixels, pixels in a uniform area of the image, and pixels with a spectral signature orthogonal to the spectral reference signal will all be completely rejected. [0095] It is not absolutely necessary that the different reference signals have to be at integer ratios of frequencies, but if they are at integer ratios of frequencies, one reference function can be constructed in real time by a continuous multiplication of the multiple different reference signals. [0096] In general, to try to find pixels with multiple attributes simultaneously, we can multiply the reference functions corresponding to the multiple attributes. To try to find pixels with one attribute or another, we can add the reference functions. [0097] In the previous descriptions “Method for adaptive spectral sensing”, “Additional comments on method for adaptive spectral sensing”, and “Method for adaptive temporal and spatial sensing and processing”, an important part of all of the methods is a mechanism (the time averaging circuit) for integrating or time-averaging the time-varying multiplied signal. There are many methods for making such a mechanism well known to those skilled in the art, including analog electrical integration or averaging circuits that return a voltage or other electrical signal, and digital schemes that would average or integrate an analog input signal after AD conversion. One of the simplest, which we mention here for completeness, is a resistor-capacitor circuit as shown in FIG. 14. This will average the input signals over a time t, in seconds of the order of the product, RC, of the resistor value, R, in ohms, and the capacitor value, C, in farads. Such an averaging time t could typically be chosen to correspond to several cycles of the time varying reference signal. [0098] Although the present invention and its advantages have been described in detail, it should be understood that various changes, substitutions, and alternations can be made herein without departing from the principle and the scope of the invention. Accordingly, the scope of the present invention should be determined by the following claims and their legal equivalents. Referenced by
Classifications
Legal Events
Rotate |