US 20060256214 A1
A method and apparatus for calibrating a sensor for highlights and for processing highlights is described. In an embodiment, a method includes identifying highlight regions in an image of a scene. The method further includes calculating flare intensity values for the image using the locations of the highlight regions. The method also includes subtracting the flare intensity values from the image.
1. A method for improving the highlight reproduction of an imaging system comprising:
capturing a multi-channel image of a scene;
identifying a highlight region in the multi-channel image;
identifying a first channel having saturated signal values in at least a portion of the highlight region;
identifying a second channel having unsaturated signal values in said portion of the highlight region; and
using the unsaturated signal values in the second channel to provide modified signal values for the first channel in said portion of the highlight region.
2. The method of
3. The method of
4. The method of
5. The method of
6. The method of
7. A method for improving the highlight reproduction of an imaging system comprising:
receiving a multi-channel image of a scene;
detecting the presence of an area of saturated values in the multi-channel image;
processing the edge of the area of saturated values to determine if the color of the edge corresponds to a color that is close to the white point of the scene; and
replacing the saturated values with values whose color corresponds to increasing intensities of the white point of the scene when the color of the edge corresponds to a color that is close to the white of the scene.
8. The method of claim 10 wherein the saturated values are replaced with values having a parabolic profile.
This application is a continuation of U.S. patent application Ser. No. 10/072,251, filed Feb. 7, 2002, entitled “IMPROVING THE HIGHLIGHT REPRODUCTION OF AN IMAGING SYSTEM” by MacLean et al., the disclosure of which is incorporated herein by reference.
1. Field of the Invention
The present invention is directed to the field of processing data from electronic light sensors that are directly exposed to a scene or that are exposed to film that has captured the scene.
2. Description of the Related Art
Photographic film is often able to capture in a single exposure the full dynamic range of a scene, which can often have a range of up to one million to one. However, this full dynamic range may not be communicated to the viewer through conventional optical and digital techniques.
Fortunately, because of the nature of photographic film and because of its wide latitude, it is possible to scan the developed film negative using an electronic photosensitive or charge coupled device (“CCD”) sensor. Although the CCD used in the scanner is based on conventional electronic sensor technology, this arrangement of film scanning can capture the entire dynamic range of the scene. The ability to capture over 16 stops (where a stop corresponds to a two-to-one increase in light level) has been demonstrated. The issue with film is shown in
CCDs and other electronic scene sensors generate image signals by producing an electron signal that is generally converted into an output voltage. CCDs for example, produce an electron charge generated by photons that impinge on the sensor area of the CCD or other electronic sensor. Consequently, this imaging process is fundamentally linear with light intensity for a fixed exposure time (and integrated light energy for a variable exposure time). Because of this linear nature, typical electronic images have a dynamic range of about one thousand-to-one. However, a scene can also have specular highlights with a range of up to one million-to-one or greater. Accurate reproduction of both the color and structure of these highlights is required if a scene is to be accurately reproduced.
What is needed and the problem to be solved is the provision of a system for increasing the dynamic range of an electronic image sensor and the ability to ensure that, for a film-based system, the scene is accurately captured over its entire range.
It is an aspect of the invention to improve the dynamic range of a captured image using calibration techniques on the image capture system.
It is another aspect to improve the dynamic range of a captured image using signal processing operations on the captured image.
It is another aspect to electronically capture a picture that has areas that are more than ten times brighter than the whitest reflective area in the picture using a highlight sensitive sensor, and to accurately reproduce those areas electronically.
A method and apparatus for calibrating a sensor for highlights and for processing highlights is described. In an embodiment, a method includes identifying highlight regions in an image of a scene. The method further includes calculating flare intensity values for the image using the locations of the highlight regions and other flare-sensing methods. The method also includes subtracting the flare intensity values from the image.
FIGS. 23 illustrates another sensor option according to embodiments of the invention.
A system for increasing the dynamic range of an electronic image is disclosed. A number of different techniques can be used to impose the dynamic range of an electronic image. It is an aspect of the present invention to electronically capture a picture that has areas that are at least ten times brighter than the whitest diffuse reflective area in the picture using a highlight sensitive sensor, and to accurately reproduce those areas electronically. It is another aspect of the present invention to improve the accuracy of the dynamic range of the captured image for both electronic cameras and film based cameras using calibration operations on the image capture system. It is a further aspect of the present invention to improve the dynamic range of the captured image using signal processing operations on the captured image. Another aspect of this invention is to ensure that in highlight areas sensors and their associated systems can provide stable, reproducible and accurate data so that the color and texture in highlights can be consistently and accurately represented as digital image intensity values. Another aspect of this invention is to ensure that, in highlight areas, sensors and their associated systems can display, in a visually realistic or acceptable way, the full captured dynamic range within the limitations of the display device. The above aspects can be attained by a system that includes highlight sensitive sensor combined with a highlight processing system. The highlight processing system can include a system for calibrating the image capture system and/or a system for applying image processing operations to the captured image.
Other aspects of this invention include: (1) calibration technique with highlights—calibrating directly looking at a light source; (2) increasing or adjusting the light level to tabulate where the switch point level of mode switching sensor systems provide greater sensitivity, to expand dynamic range; (3) identifying classes of electronic sensor photosites and apply the appropriate calibration and conversion methods to produce a faithful imaging signal; (4) outputting the maximum switch level of the sensor to enable processing to compensate for switch level artifacts; (5) using a lookup table that responds to a mode in which the sensor is operating data and applies an appropriate lookup table function for that specific mode; (6) using merged data and a lookup table with the mode as an input into image processing to correct for mode switching contours; (7) using separate lookup tables for each sensor mode; (8) correcting for sensor errors in highlight data by computing the function describing the error and then computing the inverse and applying it to the data from the sensor; (9) computing correction to be applied as a sum or product of image sensor data; (10) accurately reproducing highlights and then scaling the intensity in such a way that the texture and color of the highlight is preserved while remaining in the range desired by the film director or reproduction medium; (11) merging data from a multi-mode sensor using a lookup table; (12) performing sensor correction for highlights in two spaces, for example, a log space and a linear space; (13) automatically calibrating film by downloading the film batch or other ID on the Internet or other electronic communication system, to provide data automatically into the system for appropriate calibration; (14) updating film calibration data by uploaded film history or time since manufacture, before downloading calibration data to the scanner; using software to modify calibration data in the scanner based on film history; (15) providing digital data along with film to calibrate the film; (16) solving for the optimal function G(C1,C2) in order to ensure optimal calibration between different modes or outputs between different channels, including simultaneous output channels and outputs from single channels; (17) rendering scenes with full highlight information to match the available output dynamic range of the output device; (18) capturing the full color highlights of an image by performing calibration of the full dynamic range of the image and then processing the image in such a way that texture an highlights are preserved while ensuring that the display print or output device is operating within its available dynamic range; (19) suppressing or modifying the size or appearance of highlights; (20) modifying lookup tables based on the lens used; (21) calibrating highlights at a center or single position in a field; (22) calibrating across an entire field; (23) to prevent false color from contaminating the highlights, adding a term that reduces color saturation (by, for example, converting into YCC color space, reducing the gain on the two C channels and then converting back to the original spec, or going through a 3×3 matrix to reduce the color) to desaturate the color as the luminance increases beyond the channel saturation point; (24) extending the black range or shadow detail, paying attention to saturation and noise—noise removal; and (25) correcting the bilinear or higher order term of a lookup table that corrects for highlights to 2D bilinear or higher order, to merge multiple output channels from CCD sensors.
A system for increasing the dynamic range of an electronic image is disclosed. A number of different techniques can be used to improve the dynamic range of an electronic image. The discussion below will cover: (1) calibration, (2) signal processing, and (3) sensor changes of film-based systems using electronic scanning as well as electronic cameras. These techniques can be used alone or in combination to improve the dynamic range of electronic images.
Alternately, image data may be captured by the Electronic Still Camera (sensor & process) 4027 producing electronic data that may have been processed within the camera in numerous ways, including enhancing quality, compensating for sensor artifacts, and extending both the black and white dynamic range as described herein. The data is then passed to the Digital Processor 4022 via the Playback Memory or Direct Connect function 4028. The Digital Processor 4022 may then further transform the image so as to remove artifacts already introduced using a variety of approaches including those described herein, enhance quality using a variety of approaches including those described herein, add image information, extend both the black and white dynamic range as described herein, correct or prepare for downstream display, or print functions or perform a variety of other possible digital processes.
If the image is to be printed, it may be first stored or transmitted. This function is performed by Store/Transmit Print image data function 4023. The data then passes to Printer 4024 for printing. Alternatively, the image may be viewed in a display apparatus. To achieve this it is stored or transmitted. This function is performed by Store/Transmit Print image data function 4025 that forwards data to the Image Display 4026.
The 35 mm Motion Picture Camera 4040 uses photosensitive film to capture a motion scene. This film is processed and scanned in Film Scanner 4041 to produce digital or electronic data that corresponds to the captured motion scene. This data is then processed in Digital Processor 4042 to calibrate the image, transform the image so as to remove artifacts already introduced using a variety of approaches including those described herein, enhance quality using a variety of approaches including those described herein, add image information, extend both the black and white dynamic range as described herein, correct or prepare for downstream display or print functions, or perform a variety of other possible digital processes.
Digital Processor 4042 may also change the artistic look of the image. The concept of look is associated with terms such as “film look,” “soft look,” or “harsh look,” and is based on the assessment by artists of the appearance of the image. Manipulating a variety of image properties including tone scale, sharpness, noise, and image grain changes the look.
Digital Processor 4042 may include conventional digital post-production processes such as digital intermediate processes, similar processes referred to in Brad Hunt, Glenn Kennel, LeRoy DeMarsh and Steve Kristy “High Resolution Electronic Intermediate Systems for Motion Picture” J. SMPTE, 100: 156-161, March 1991 or the digital functions performed by companies such as Cinesite®. The processing may also include the functions of Electronic Preparation of Elements, Color Correction layers, or the various conversion functions referred to in Robert (Bob). R. M. Rast, “SMPTE Technology Committee on Digital Cinema—DC28: A Status Report” J .SMPTE, 110: 78-84, February 2001.
Instead of using film, image data may be captured by the Electronic Motion Camera w/ digital process 4047 producing electronic data that may have been processed within the camera in numerous ways including to transform the image so as to remove artifacts already introduced using a variety of approaches including those described herein, enhance quality using a variety of approaches including those described herein, add image information, extend both the black and white dynamic range as described herein, correct or prepare for downstream display, or print functions or perform a variety of other possible digital processes.
The in-camera processing may also change the artistic look of the image. The concept of look is associated with terms such as “film look,” “soft look,” or “harsh look,” and is based on the assessment by artists of the appearance of the image. Manipulating a variety of image properties including tone scale, sharpness, noise, and image grain changes the look.
The in-camera processing may include conventional digital post-production processes such as digital intermediate processes, similar processes referred to in Brad Hunt, Glenn Kennel, LeRoy DeMarsh and Steve Kristy “High Resolution Electronic Intermediate Systems for Motion Picture” J. SMPTE, 100: 156-161, March 1991 or the digital functions performed by companies such as Cinesite®. The processing may also include the functions of Electronic Preparation of Elements, Color Correction layers, or the various conversion functions referred to in Robert (Bob). R. M. Rast, “SMPTE Technology Committee on Digital Cinema—DC28: A Status Report” J .SMPTE, 110: 78-84, February 2001. The data is then passed to the Digital Processor 4042 via the Playback HD Tape or Cable Connect 4048. The Digital Processor 4042 may then further calibrate the image, transform the image so as to remove artifacts already introduced using a variety of apparatuses including those described herein, enhance quality using a variety of apparatuses including those described herein, add image information, extend both the black and white dynamic range as described herein, correct or prepare for downstream display or print functions, or perform a variety of other possible digital processes.
Data from the Electronic Motion Camera w/ digital process 4047 is then passed to the Playback HD Tape or Cable Connect function 4048 and then passes to the Digital Process 4042 function that performs some or all of the aforementioned functions
If the image is to be passed to film, it is first recorded to film and then, in the preferred embodiment the film, is duplicated using film printers common in to the motion picture industry. This function is performed by Print to Film 4043. The film then passes to the Film Projector 4044 that projects the images to the motion picture screen.
Alternatively, the image may be viewed using an electronic projector. To achieve this, data from Digital Processor 4042 is stored, transmitted, or both. This function is performed by Store/Transmit 4045 that forwards data to the Electronic Projector 4046. The Electronic Projector 4046 projects the images to the motion picture screen.
Multi-mode electronic sensors, as disclosed in this invention, are one way to capture a wide dynamic range scene. These multi-mode sensors have modes with different levels of sensitivity. As a single image signal is generally required by downstream electronics processes, it is necessary to merge the signals from the each of the modes.
One type of multi-mode sensing system includes normal charged coupled device (“CCD”) elements and less sensitive CCD elements that only reproduce the highlights. These elements can be rendered less sensitive by being smaller than the primary elements or having a shorter integration time. This is shown in
The method of communicating which one or whether both CCD wells contribute to the output signal can be as described below.
During output of the CCD signal the output of both CCD elements can be sent to the output of the CCD chip. Alternatively a threshold sensor associated with CCD well 101 can sense whether it is at or near saturation and if so the output signal from CCD well 102 can be inserted in place of the output of CCD well 101. Along with this output would be a signal indicating which CCD (101 or 102) is responsible for the output signal. Although the signals within light sensors, associated circuitry, and connectors may in fact go either positive or negative with increasing light depending on the specific design, for ease of understanding descriptions herein, the convention that increasing light produces increasing signal is generally used.
further method is to output the signal from both CCD wells. It is also possible to output the signal from the appropriate CCD well and indicate whether the signal is derived from 101 or 102 by changing the polarity of the output signal with respect to a reference level, for example signals above the reference level would correspond to 101 or signals below the reference level would correspond to 102. It is also possible to change the reference level to indicate whether it is from 101 or 102.
One method of implementing these types of techniques is shown in
Simultaneously inverter 204 generates a signal to gate transistor 205 which is normally the operative transistor to transfer the charge from CCD site 101 and disables this so that the charge to the CCD shift register on line 206 can come from either CCD site 101 if it is not full, and from CCD site 102 when 101 is full or is close to full.
It will be appreciated that these operations could be undertaken multiple times with different periods of charge integration during the scene exposure time. These periods could be in a binary length of time in the sequence 1, 2, 4, 8, but are more likely to be in higher ratios such as 1 time increment, 60 time increments, 3600 time increments for a total integration time of 3,600 time increments (using this technique the total integration time is equal to the longest integration time, since shorter integrations occur during the longer integrations). Integrations however can be sequential leading to an integration time of 36,061 time increments or longer.
The charge signal from the sites can be stored in a memory 408 (see
Another type of multi-mode sensing systems uses a dual integration time approach where a short integration time is used along with a longer integration time. The short integration time is first discharged, then the longer integration time is discharged from the CCD, and then the two images are combined.
This is shown in
At the end of the primary integration time 302 and during period shown by waveform 313 the main discharge cycle takes place. This discharge cycle is subsequent to the discharge cycle during period 307 and can include the charge contribution of all cells whether they were predischarged or not.
During the second discharge cycle 313, the signal on line 410 stored in memory 408 is read out on line 410 in synchronism with the discharge data coming out of CCD 401. If the signal 410 indicates that the currently discharged CCD cell site has already during the 307 cycle outputted discharge data (due to the fact that it was in a high intensity region of the scene) the memory 408 control circuit 411 is signaled to prevent a further write cycle of the data on line 407 through memory control lines 412 consequently the original data in the memory is left undisturbed, but if the information of the second cycle is preempted by the discharge of the CCD during the first cycle, then new data is written into the memory. It will be appreciated that in order to achieve a wider dynamic range these cycles may be performed repeatedly.
Calibration can be performed to correct for a number of different aspects of the captured electronic image to improve the dynamic range of the image. Sensor calibration ensures that the full dynamic range of the sensor is utilized. Color calibration ensures that the optimal spectral sensitivity is utilized. Artifact calibration reduces the deleterious affect that artifacts have on the dynamic range of the image. Each of these types of calibration will be discussed below.
Meeting the improved dynamic range challenge requires an accurate calibration, over all sources of variability, of the photo sensing system response. This may be done once for each light sensor design, may need to be done for every sensor manufactured, or may need to be done for each film type or for every batch of a given film type. The decision of how frequently these calibrations need to be generated depends in part on manufacturing variability and device stability due to time, temperature, and other variables. The sensor system must also be tested for other sources of variability, such as flare, that must be characterized as part of the overall sensor system model produced by the calibration process.
Color and attenuator wheel 210 is positioned between uniform light source 204 and camera 202, and is rotatably coupled with shaft 212 such that color and attenuator wheel 210 is able to be rotated in front of uniform light source 204, presenting a different color or image brightness and sensitivity filter through which uniform light source 204 can cast light onto camera 202. The intensity of the light can be measured in a separate photometer (not shown). The color and attenuator wheel 110 contains color filters and neutral density attenuators. Each filter or attenuator spans the region between spokes of the wheel so that the light color or brightness is modified and uniformly illuminates the sensor.
For an electronic camera, specific density values of color and attenuator wheel 210 are chosen so that the full dynamic range of the sensor is calibrated by exposing the sensor to light values that fall across that range. In addition, if a break-point approach or multi-mode approach were chosen for light sensor 206, then the exposures of color and attenuator wheel 210 are particularly concentrated at the intensities where a mode change occurs, so that the key data values that need to be used to establish correct calibration are produced. In the case of nonlinear approaches, such as those used in some CCDs (or in film), the values would be distributed across the dynamic range in such a way as to accurately measure both the linear and non-linear aspects of the response. If it is not necessary to compensate for characteristics of individual pixels in the light sensor 206 in the case of an electronic camera, then not all of the CCD pixels need to be exposed or analyzed.
To control the spectral illumination and to limit heating, an IR filter 208 (illustrated in
It is also possible to use a mask 230 in close proximity to the light sensor of camera 202 to image light source 222 as a shadowed highlight in small areas of the sensor to prevent the light sensor from being affected by the average picture level, or suffering from temperature changes from the intensity of the light. Camera 202 may then be shifted in XY to ensure that all pixels view a uniform highlight area during the calibration process.
Having been fully calibrated, a signal is produced on RGB lines 258. These signals pass to local area tone scale correction functions 259 that scale the tone of the highlight to accommodate the available dynamic range from the camera or scanner output 260. These processes may be performed in other sequential orders in addition to the order laid out here, and techniques disclosed herein may perform multiple functions within one technique.
When a highlight occurs in a scene, it is as if the sensor was pointing directly at the light source that is illuminating the scene. Normally, an image is formed by diffused reflection from objects, and so the amount of light that reaches the sensor is a very small fraction of the total light from the illuminator. In areas of highlights, this fraction is much larger, and so the intensity of the light may be up to one million times brighter than the more common diffuse areas of the scene. To simulate these higher intensities, special methods are required to generate highlight calibration light sources.
These special methods and equipment that provide highlight calibration light sources will now be discussed.
Light sources 610 and 612 provide uniform specular highlight areas, and their intensity may be calibrated as a function of the intensity of the light and the diameter of diffuser 620. Light sources 610 and 612 may also be partially blocked to provide multiple levels of intensity. By choosing light sources with different intensities of lamps and having different sizes, it is possible to obtain highlights covering a wide range of intensities. Because this arrangement allows multiple highlights in each reflective surface 604, 606, 608, calibration may be expeditiously performed, as each scene will contain multiple highlights from each of the multiple domes. Furthermore, knowing the reflective characteristics of the domes, accurate information may be obtained about the intensity of the highlights. In addition, in this scene it is advantageous to include a diffuse gray scale and color target within the test image 602 so that the light sensor is calibrated across the entire dynamic range of the sensor.
If other variables significantly affect the response of the sensing system, it is necessary to ensure that the highlight images are captured across a range of those variables that is likely to be encountered. If individual photosites produce different responses, as can be the case with some electronic sensors, it becomes necessary to mount the light sensor in camera 202 on an X-Y stage, such that light sensor 206 may be moved at right angles to the optical axis of lens 214 (while remaining in focus), and such that every photosite is exposed to every highlight combination. If necessary, during this process, data may be collected regarding highlight and overall response of every light sensor as a function of the highlights and patches viewed and also as a function of other variables. This large database constitutes the raw sensor calibration data in response to test scenes including highlights. It is necessary to ensure that this database contains the information regarding the specific photo site, highlight, and patch response, as well as other variables for every sensor response collected.
To achieve this indexing information, which links highlight response to a specific photosite or device, it may be desirable (from a productivity point of view) to use image processing or analysis software that, based on knowledge of the scene X-Y, conventionally coordinates size, spectral intensity, or color of every highlight. This software generates a database that tabulates light sensor response to those variables that may have a significant effect on the response.
The testing fixture of
The artifacts of flare, scatter, and halation have the effect of causing highlights within an image to influence other areas of the image where the highlight is not found. These artifacts can have spectrally varying responses, e.g., a white highlight can scatter red or blue light into other parts of the image. Sometimes, this scattering of additional light occurs across the whole image; for other artifacts, it is more pronounced in the immediate vicinity of the highlight, and does not affect parts of the image that are more distant from the highlight.
These artifacts can arise from a number of factors, including lens design or manufacture, light scatter within the lens support mechanism, lack of sufficient shielding of the lens, stray light bouncing off the sensor and then back onto other parts of the sensor, or when film is used to initially record the image light scattered from the back of the film that is sensing the image, or from scatter or light piping within the layers of the film due to scatter by particles such as silver halide particles or other particles within the film. Similar effects can also occur from the diffusion of chemicals during film processing, or from chemical starvation and similar mechanisms.
As already mentioned, to characterize these effects, calibration techniques such as those shown in
When responding to image areas of high intensity, image sensor systems, whether they be film-based or electronic-based, can change in their spectral sensitivity response as a function of scene intensity. These changes in spectral sensitivity response may occur as a result of a variety of reasons, including non-linearities within the sensor, the effect of halation, or light scatter within the sensor or saturation effects. Any sensor that is inherently non-linear can change its spectral characteristics with regard to intensity, especially if there is a range of narrow spectral band intensities in the scene. Calibrating and correcting for these effects cannot be performed simply by using traditional linear methods such as 3×3 matrices to convert from the sensor color space into the specified output color space in digital data of the sensing system. At a minimum, when using 3×3 matrices, in the color conversion, it is necessary that the 3×3 matrix operators have coefficients that change as a function of the intensity of the image in each channel or as a function of the overall image intensity at the pixel where the calculation is being performed. Thus, in order to produce calibrated color output channels, each channel must be computed using color conversion from color source channels.
In normal operations, these input coefficients are fixed for a given color conversion; however, in areas of highlight or in dark regions, experiments concerning color accuracy with intensity can show that more accurate color rendition is achieved if these coefficients change according to experimentally or analytical derived values. In addition, to accurately handle color in high intensity or in deep black areas where the sensor is non-linear, it is necessary to use additional computing methods based on the calibration data previously discussed. For example, based on data from channels 1504 through 1508 (and possibly 1510), it may be necessary to use a lookup table or other device to compute the necessary deviations from nominal fixed values of the input coefficients 1520, 1522, 1524, and possibly 1526. Other approaches to address this problem of non-linearity include the use of three or higher-dimensional lookup tables to perform the function shown in
The values of these corrections can be determined using methods such as inverse polynomials, Fourier functions, numerical correction methods or other methods described herein or referenced. These methods use the experimental or analytically derived calibration data. These calibration techniques are able to provide both the necessary raw data for both the non-linearity of individual color channels and the modification of the spectral sensitivity of these channels. However, it may be necessary to know the spectral characteristics of the calibration light sources referred to in other parts of this disclosure, to accurately determine the required correction that transforms the color channels to a calibrated output color space.
A multiplicative and offset correction for the shading and flare may be applied using techniques outlined in U.S. Pat. No. 4,051,458 (Video amplitude related measurements in image analysis Morton, Roger R.) or by determining across the lens the for each pixel, multiplicative and offset correction values, which when applied to the image data will, remove shading.
The correction for the measured variation is applied (336) across the field to all pixels of light sensor 206. producing image data that is corrected or compensated for lens shading. This method corrects for both average center lens/light and sensor effects. Next, light sensor 206 is typically illuminated through the lens with highlight test objects as shown in
Once the data from the calibration processes described herein are generated, it must be translated into either lookup tables or functional expressions, such as polynomials or Fourier functions, so as to apply correction operations to the data produced by the electronic sensor. Alternatively, a combination of lookup tables and functional corrections may be used. The decision regarding the appropriate combination of correction functions is made based on considerations of overall accuracy, throughput rate, and cost.
The generation of lookup tables and functions is well known in the art. The process includes multilinear (cubic, quartic, or higher) interpolation techniques, polynomial techniques, techniques for defining spline functions, Fourier transforms, and other transform approaches. Also, least squares error minimization methods may be used. In such methods, a function may first be defined to fit the data from the calibration process. This function may be a polynomial or other function that describes the electronic image sensor response.
The general formulation of these calibration approaches is as follows. A response function 0 corresponds to the value of sensor image data at a defined point in the system. This may be an array of data derived from calibration processes herein described, and is defined as
This function describes the measured pixel-by-pixel (or group of pixels such as a 128×128 subsample of the image) image data response from the calibration of the sensor, where X1, X2, . . . , Xi, . . . , XN are the inputs affecting the specific pixel X0. The variables X1, X2, . . . , Xi, . . . , XN correspond to highlight intensity and color of the highlight, which are all imaged to the pixel X0. Other variables such as temperature and overall light intensity (including those mentioned previously) may also be assigned a variable such as Xj and included if it is found that these variables have an effect on the electronic image sensor response, such that the exclusion of these variables would result in degrading the calibration accuracy to an unacceptable level. Adding additional variables increases the time required for calibration and the quantity of image data required to populate the array. The analytical form of function O termed f(X0, X1, X2, . . . , XN) may be computed using least squares methods to determine the coefficients of the function. In general, if the function is fit to a polynomial, there will be a vector or matrix C corresponding to the coefficients of the polynomial that describe the image data calibration response from the sensor. In practice, the number of coefficients in C need not be so large if it is found that, for example, pixel response to some variables can be described in a linear manner while other variables elicit a higher order response. The least squares method involves comparing the square of the error between the analytical response of the electronic sensor defined by f(X1, X2, . . . , Xi, . . . , XN) and the actual response O=F(X0, X1, X2, . . . , Xi, . . . , XN) as recorded in the data array generated by one or more of the calibration processes described above. This error is summed over all data points, and then least squares optimization methods are applied to find the optimal set of coefficients C that produce the minimum sum of errors. Other forms of the function f may include tetrahedral, bilinear, cubic, or multi-linear approximations.
The least squares method may be implemented by computing using Newton, Conjugate, or related techniques. Correction functions correct the image data response of the sensor for errors introduced into the image data. As described herein, correction functions are implemented using lookup tables directly addressed by pixel data values or by using arithmetic computational approaches of applying corrections to the image data. These corrections may be driven by parameters from lookup tables or from coefficients implementing analytical correction functions such as polynomial correction functions. Whichever approach is used the correction function applied to the image data from the sensor can be defined analytically as function k with coefficients M1, M2, . . . , M. The correction function may be represented as or be a data array K or as a function k. Either correction function K or k may be derived using error minimization where the minimal error defined as the sum of the pixel errors across a calibration image data from the sensor, where the pixel errors are defined, for each pixel, as
A number of approaches for computing solutions are possible:
Once the data set K has been computed, it can be imbedded in a lookup table or the function k may be converted to look up tables that drive arithmetic computational techniques or approaches of applying corrections to the image data.
An alternative method to the least squares approach is to compute, an inverse G to the function f under the conditions imposed by function D. This inverse is calculated to give desired values V1, V2, . . . , Vi where V1, V2, . . . , Vi corresponding to the desired sensor image data. These desired values occur at the defined point in the system in response of to the calibration images herein described. Thus, this inverse G(O, X0, Xi+1, . . . , XN) applied to the sensor image data will now provide correction to the image data all values of X1, X2, . . . , Xi. Therefore, by feeding the electronic sensor response into this function, actual scene corrected image data will result. There are methods other than the least squares method for computing function k from K, including Fourier and local neighborhood methods, such as multi-dimensional bi-linear and bi-cubic interpolation, tri-linear and tri-cubic interpolation, or multi-linear and multi-cubic interpolation methods, as well as other polygon-based methods. These methods determine the equations describing a surface or function for a local neighborhood of data points, that when applied to the image data produced by the calibration processes described above, result in a the desired calibration response. When applied to scene image data the result is corrected data that more accurately corresponds to the scene. The implementation is shown in
Thus, the data on sensor output line 406 becomes the addresses to lookup table 414. Other address data to lookup table 414 include the data on calibration data line 410 to produce on corrected scene data line 412 the output data from lookup table 414 corresponding to the desired result of V1, V2, . . . Vi.
Lookup table 414 may be programmed by writing into the lookup table the function 408. This is done by programming a computer with the function 408 according to correction function K or k and then arranging for the computer to generate the appropriate control signals to a device that is designed to program lookup tables or memory. The specific device will depend on how the lookup table is implemented. Implementation techniques include read only memory (“ROM”), flash memory, programmable read only memory (“PROM”), field programmable gate arrays (“FPGA”), random access memory (“RAM”), or magnetic disk drives operating with a processor and possibly including a cache to improve response time.
Another way to implement the function 408 is to use a digital signal processor (“DSP”) to implement the function directly. This is done by programming the DSP with the function 408 such that, as data enters from the sensor 402 into the DSP, it computes the specific values for function 408 based on the input data. Another way to compute function 408 is to directly use a least squares method or other optimization method in a manner similar to that described previously for determining the function K. This approach avoids the need for taking the inverse of F, while giving the result as correction function 408.
The actual implementation of the correction of electronic image sensor data is often independent of the way in which the correction function is generated. Thus, any of the techniques mentioned above may be applied to lookup tables or software implemented in polynomial or multilinear interpolation techniques.
In many cases, the lookup table interpolation technique just described can produce very large tables that are not economical. For this reason, it is appropriate to take steps to reduce the size of lookup tables. The methods for doing this include:
In addition, other methods may be used to reduce the size of the lookup tables.
Often, it will be necessary to correct for the non-linearities of individual pixels on a pixel-to-pixel basis. However, building a one thousand word lookup table for three million pixels or more, coupled with additional banks of tables for various parameter defined conditions, may be economically prohibitive. In this situation, one approach is to build 100 or 1000 lookup tables, each 10- or 12-bit table corresponding to a particular class of pixel response. It is then necessary during the calibration process to classify the pixels, possibly using clustering techniques, building, for example, 100,000 pixel-type lookup tables and then identifying for each pixel the lookup table that is most appropriate. In this case the relation of the appropriate lookup table is driven by a pixel classification table that receives the current pixel number or address the appropriate classification, as is shown in
Clustering pixels according to the correction required takes advantage of the fact that, although there may be millions of individual pixel sites in the sensor or sensors used in a camera, the correction required for each one may not be so different, and the same correction may be applied to multiple pixel sites. The technique for implementing this using a look up table directly addressed by pixel data values is shown in
Once function 408 is determined as described above for every pixel, a matching process is used to identify those pixels that produce the same correction response. One way to do this is to select an initial pixel, identify its correction response, and compare the response data of all the other pixels to find pixels with the same or similar correction response. Once these pixels are identified, they are considered a cluster having the same type of response, and are identified as having been selected. This process is repeated for all other unselected pixels within function 408. Once the point is reached where the remaining clusters have few pixels, these remaining clusters are grouped with existing clusters if the resulting error is acceptable. In cases where the result is not acceptable, they are considered separate clusters. Thus, pixel type lookup table 508 is generated as an index, indicating to which cluster each pixel belongs, and lookup table 512 is simply the correction for each cluster type. Thus, the pixels are grouped into pixel types by cluster.
This technique can group all of approximately six million pixels in an HDTV 3-sensor camera into 10,000 types, producing a lookup table size reduction of the order of 100 to 1,000 fold. An alternative clustering approach is to break clusters by the image data output signal and add additional inputs from image data line 504 along additional input line 516, to the pixel type lookup table 508.
Using look-up tables that drive interpolators that may be linear, spline based on polynomial involves using a lookup table that is not as densely populated as the data requires but rather is populated, at every 100th point (in general every Nth point) across all of its dimensions. Thus, if the table is a four-dimensional table and it is populated at every 10th point along every axis, then there is a savings in lookup table capacity of 10,000 to 1. Thus, at a desired data point, the points surrounding the desired data point are accessed (in the case of linear interpolation, this corresponds to two points, bilinear—4 points, trilinear (or linear interpolation over 3 dimensions)—8 points, and four-dimensional—16 points). Data is interpolated between pairs of points using straight-line interpolation, and these interpolated points are then again interpolated until the value at the desired point is reached. Cubic interpolation may also be used, in which triples of points instead of pairs of points are accessed. Thus, this approach generates the correct value from surrounding values by computing from the surrounding points, in real time, the interpolation coefficients to reach the desired value. In linear interpolation, these coefficients correspond to the difference between pairs of points that may also be written as the equation
This is shown in
An alternate approach of generating correction data for existing image data is to have the lookup table not compute the final values, but instead compute values that, when added to or multiplied by the output from the sensor (or a combination of both for higher order functions), produce the desired result. This approach is attractive because if the output from the electronic image sensor is already in the color space or units for which this approach is most effective, then the implementation can be quite simple. In most cases, the most appropriate space in which to do this is a linear sensor space—one whose output values are linear with light intensity; however, one disadvantage of this approach is that it may require a greater number of binary bits of precision. An embodiment is shown in
The merged output from light sensor 1602 is then transformed in space transformation function 1606 into a space where corrections can be added or multiplied by adder 1608 and multiplier 1610, which also receive inputs from lookup table 1612. The inputs from lookup table 1612 may be controlled by sensor data line 1614 using data directly from the sensor, from merged data line 1616, or data in the chosen correction space from correction data line 1618. It is also possible to perform correction in two spaces, for example, a log space and a linear space, in which cases functions such as adder 1608 and multiplier 1610 would be performed in a linear space with intensity.
Then, the output on output line 1620 would be transformed into a logarithmic space and as shown in
Using hardware polynomial calculators driven by coefficients set by look-up tables is a technique that involves using hardware-or software-implemented mathematical functions. These mathematical functions may take virtually any mathematical form including linear, spline-based, discrete cosine transform (“DCT”), wavelet, Bessel function, or polynomial. One advantage of discrete cosine transform is that application-specific integrated circuits (“ASICs”) are already available to implement this function.
The mathematical function used in function 1706 may have the same form as the correction function or may be transformed from function 1408 into some other function as part of the calibration process. Calibrated and corrected scene data is the result of the mathematical function 1706 and is transmitted on calibrated and corrected scene data line 1714.
Another way to reduce lookup table size is by compression methods such as Lempl-Ziv and other methods. These require that the data be stored in the lookup table in compressed format and that there be a decompression function at the output of the lookup table to transform the data to its desired decompressed format.
Many of the techniques described above may be used in combination to effect an even greater reduction in lookup table size. The decision of which to apply and in which combination is largely an issue of economy, throughput, and power consumption.
As shown in
A function for the lens may be determined by a1(x, y). The red, green, and blue response of the lens and the overall system response may be estimated from the sum of
Because of the extent of the flare function a(x, y), it may be computationally intensive to use the convolution method. The flare image shown in
Data from highlight cells need only be outputted when one of the R, G, or B cells or more is saturated. Thus, in the output stream the signal from the highlight cells can be substituted (with an appropriate signal) for one or more of the saturating cells. This type of sensor produces data on a single line and the data is as shown in
In cases where a sensor may saturate when encountering a highlight, signal processing strategies associated with detecting the highlights may be applied. For example, because there are two classes of highlights, i.e., glossy highlights and metallic highlights, the edge of the highlight must be discerned. Highlights from glossy surfaces are characterized by a rapid change in color from the diffuse reflection color of the surface to the color of the light sources. In an embodiment, the method includes detecting the presence of a highlight using image processing algorithms that detect saturation. In a correctly exposed scene, the presence of a highlight will correspond to an image data signal at the saturation level of the sensing system, and provided that the saturation level of the system is above 100% to 150% diffuse white, corresponding to a highlight or light source in the scene.
The method further includes processing the edge of this saturated area to assess the color in the immediate surround. This may be an area only a few pixels wide around the highlight. If this surround has a color point that corresponds to the white of the scene or is close to the white of the scene then this area is assumed to correspond to a scene light source or a glossy reflection. The method also includes replacing the saturated values with pixel values whose color point corresponds to increasing intensities of the white point of the scene.
This is shown, for example, in
Dynamic range can be extended by using the data from the non-saturated color channels. Typically, in color CCD imaging systems, one of the color selective channels will saturate before the remaining channels. This channel saturation sets the maximum usable dynamic range of the device. For the purposes of highlight reproduction, the color information may be less important than the tone scale information. This is because, in a scene with a single color illuminant, there can be two limiting cases of specular highlights. In one case, the highlight takes on the color of the illuminant. In the second case, the highlight takes on the color of the object. The remaining cases are often a blend of these limiting cases. Either reproduction method is adequate for extending the dynamic range to an extent limited by the color of the highlight. For example, if the highlight is white then the color channels may saturate at a similar level, thereby limiting the amount of extension available using this method. Alternatively, the sensitivity of different channels may be purposely mismatched in sensitivity.
In an embodiment, the regions in which channel saturation occur are identified. The neighboring color is analyzed. The saturated channels are interpolated assuming constant hue. To prevent false color from contaminating the highlights, a desaturating term is added to desaturate the color as the luminance increases beyond the channel saturation point. This provides an effective model for highlight reproduction. The highlight information is reproduced using extrapolation, taking the region surrounding the saturated region into account. The color and luminance slope may be determined. The data in the saturated region is presumed missing. The missing data is reproduced using standard curve fitting techniques. The curve fitting techniques that generate smooth data fits are preferred. The desaturating term is added to desaturate the color as the luminance increases beyond the channel saturation point. Desaturating reduces color saturation (for example, by converting into YCC color space, reducing the gain on the two C channels, and then converting back to the original spec or by going through a 3×3 matrix to reduce the color) to desaturate the color as the luminance increases beyond the channel saturation point.
Another method of detecting highlights is to give up part of the available dynamic range of the output device and apply the highlight using an electronically generated single tone scale transformation as shown on
Another alternative is to pursue a step-by-step processing approach. This method comprises knowing the full dynamic range of the target device or knowing the calibrated standard signal data values that will fall within the dynamic range of a device or group(s) of devices. The method further includes establishing, from an artistic point of view, the way in which highlights are to be rendered. The method also includes sensing highlights within the image and rendering them artistically. While the knowing and establishing may be optional, the essential step of rendering the highlights must be performed if highlight information is to be preserved. Fortunately, the human eye can perceptually interpret highlights correctly even if the highlights do not match their original brightness. Thus, the detailed structure and color of a highlight can be inserted in an image at a brightness that is much lower than the value in the scene, and the image will still appear realistic and essentially carry the full information.
Highlight sensing within a region is shown in
Transformation 1176 that continues upward corresponds to the situation where no transformation occurs. Specifically, input range 1178 corresponds to the normal input range of a scene without highlights, while the output range 1180 corresponds to the available dynamic range of the output channel or output device. Thus, the input data on highlight signal line 1102 will correspond to the output data on modified tone scale line 1110. However, as a highlight, for example, one which has ten times the brightness in scene exposure space, occurs corresponding to line 1152 in
The shape of the transformation curve affects the look and artistry of the image. For example, transformation 1188 corresponds to a transformation having the same highlight attenuation ratio as transformation 1174, but with a different shape giving a less contrasted look. As discussed previously, the lookup table, which is a function of input value, distance from a highlight, and intensity of a highlight, can be very large. However its size can be reduced using the techniques already discussed, including tri-linear interpolation. As already mentioned, such a lookup table may handle colors in the same manner or be sensitive to the fact that highlights can be colored, and generate tone scale transformations for each color channel, for example red, green, and blue.
Another method for handling highlights within the dynamic range is to suppress the highlight by reducing its size. This technique senses the presence of a highlight and then substitutes for the highlight values of surrounding colors. This approach may first identify whether the highlight is from a glossy or metallic surface. The approach for glossy surfaces can be to substitute the value of immediately surrounding pixels, while the approach for metallic surfaces may be to suppress the brightness of the highlight while preserving the color and texture.
This operation is further explained in
One approach to estimate the level of 100% diffuse reflectivity is to histogram each of the color channels across the scene. For each of red, green, and blue, the method involves plotting the value of intensity of the red, green, or blue along the X axis and the number of pixels at that intensity level along the Y axis. The level of the 100% white point can be chosen as the range at which the histogram count drops off for each color. There will be other non-zero values of the histogram count at brighter levels. For correctly exposed scenes, the point at which the histogram count drops off will fall within a range of the red, green, or blue level. Thus, the points at which the histogram count drops off for each color define the white point of the scene.
Different image applications will require different levels of image quality. Therefore, for various formats such as television, DVD, cinematographic projection, image preview, print, and graphic arts, the same degree of dynamic range, image fidelity, and accuracy may not be required. Thus, different degrees of processing, as well as additionally intercepting the image data stream and outputting for certain applications, may be appropriate.
Faithfully capturing the dynamic range of an image is an important part of preserving overall image quality and maximizing the dynamic range of an electronic image; however, many display and printing devices are unable to reproduce the up to million-to-one dynamic range that can be found in an accurately captured image. To ensure that the faithfully captured image lies within the dynamic range of the output device, regardless of whether it is a cathode ray tube, LCD display, ink jet printer, photographic print, television set, or movie screen, it is necessary to process the signal appropriately.
The method of processing the database collected during calibration depends on the method used for processing image data from the sensor. There are various ways to extend the range of a CCD and other electronic scene sensors. These various sensors may be grouped as follows:
Depending on the type of output, different correction strategies are applied.
The output from the class of system that includes a single output at multiple discrete sensitivities may take on one of several values, depending on the level of sensitivity. The goal of processing this data is to ensure that seamless transitions occur between data from one mode and data from another. This is extremely important because discontinuities between modes can produce contour lines in the resulting image produced by the sensor system. It is also important that the sensor not have a gap in values between the two modes. Alternatively, if there is a gap, it is important that it be minimized.
Another technique for smoothly handling the transition between modes is to use sensors that simultaneously output data from both modes to form a weighted average between the two modes. For example, in a region of increasing intensity, the weighting of the signal between the two modes changes from a more sensitive mode to a less sensitive mode, before the less sensitive mode saturates. Thus, the weighting factor is programmed into a lookup table and is a function of light intensity. Another approach for a single output is to apply a digital processing function (similar to the feathering and rubber stamp operations found in Adobe Photoshop version 6.0) on the resulting image in the vicinity of the amplitude, where transition between modes are to occur.
If the sensor is of the type that requires calibration correction for each pixel, then the pixel address may enter lookup table 1802 on line 1812. This may substantially increase the size of lookup table 1802. Similarly, if other parameters such as temperature, overall image brightness, flare, response to neighborhood pixels, latent image keeping, sensor age, etc. also influence sensor response, then the size of lookup table 1802 will also increase. These variables may also be received on line 1812, and lookup table 1802 must be constructed (based on either calibration or modeling of the sensor) to generate values that are appropriate under those varying conditions, and based on this parameter data received on pixel value line 1812.
Some of these secondary corrections may not be so severe as to require an expansion of lookup table 1802. They may instead be entered as additions or multiplications of the lookup table output, and therefore depending on the overall system design as shown in
One of the key issues in this type of system is managing movements in the intensity level thresholds between modes. Errors between the points where the lookup table expects the switch and where it actually occurs may, even with the image processing outlined above, introduce significant image artifacts. As previously discussed, a thorough calibration approach can address this issue, provided that all variables are monitored.
To address these errors, a number of techniques can be used. One is to monitor the switch point within the sensor by having the voltage switched in the CCD or light sensor so that, at regular intervals, the threshold level at every photo site is output in a data stream. This threshold stream is stored in a separate memory. This threshold memory data is then applied to the lookup table 1802 on line 1812, and lookup table 1802 is configured to correctly compensate for those levels. This is shown in
Another method of managing the threshold point is to output it as part of the image stream together with the black level. This level would then be sampled by a sample and hold circuit. This sample would be used as a reference whenever the highlight or high intensity mode is enabled. This is shown in
These levels may be different when outputting during the high intensity mode. For example, if the sensor operates at a sensor threshold value, then the level on threshold sensing level 1906 may be the threshold value at which the decision is made to prematurely discharge the photo site during a first phase discharge cycle. In this case, the appropriate white level is this reference voltage V multiplied by the ratio of time when the sample is taken to the total possible integration time. This ratio is greater than one and will be a reference to the black level. Thus, in this case, a gain amplifier 1942 (shown using dotted lines) may be inserted, and its accurately controlled gain sets the reference of A/D converter 1920. Using a dual photosite, this gain must be compensated by the ratio of the sensitivity between the sites or the gain may be embedded in to the data used to create the lookup table.
The techniques explained below for multiple output channels for each color may also be used with a single output for each color at multiple discrete sensitivities and other imaging methods. To use it with this method as shown in
To process signals from sensors that provide a single output for each color with a non-linear response to intensity, so as to achieve the necessary dynamic range to handle highlights, the techniques shown in
Scanned film systems where the initial image is on film and the final image is electronic fall into this class. Calibration may be performed using the techniques shown in
The non-linearity of film in a film to electronic image system may be similarly calibrated using an automatic system for calibrating film by downloading the film batch or other ID via the Internet or other electronic communication technique to provide data automatically into the system for appropriate calibration. This is shown in
Additional data such as time of exposure, conditions of exposure, film development methods, and related data are also sent by user system 2012 to data server 2010. Based on this information, processor system 2020 can modify the calibration data stored in storage unit 2006 so that the calibration data is appropriate for the conditions communicated from user system 2012. This calibration data is then loaded into user system 2012 to ensure that, during the process of scanning the film for which the calibration data has been provided, the data produced is accurately calibrated to the film.
Another method for ensuring that the film scanning process is accurately calibrated to the film is to send the data along with the film in an appropriately encoded form. Thus, the data could be stored in a CD ROM or bar code system associated with the film, and the user system scanning the film could read the information at time of scan. Another way to modify data based on film history including processing and developing is to perform the modification functions within the scanner based on the data downloaded directly from encodings on the film.
Different approaches between these two extremes may be used. For example, data may be provided to all regular users of scanners in the form of a CD ROM, and this may be used as the basis for calibration using functional or lookup table approaches. Data, which is batch-specific or process-specific, may then be downloaded as a modification for the base data. These modifications may be incremental fractional, first, second, third, and/or fourth order changes to the data or may be parametric where the data may be stored and transmitted as coefficients of a pre-established correction equation that is applied to the image stream to correct for film characteristics. The parameters or coefficients of the equation may be re-specified or incrementally modified by the downloaded data.
When multiple outputs for each color are provided due to for example, mode switching in electronic sensors, multiple data streams are produced from each output, or a mode status signal may be used to show an output mode that is valid for each of the light levels encountered by each photo site or outputs from each channel may be available. When multiple output channels are used for each color, it is generally necessary to merge these multiple output channels into a single channel, or to provide separate images or image layers corresponding to each channel. However, if separate images or image layers are used, calibration can be on a per image basis. Generally, it will be necessary to merge these various image layers, and this merging operation of calibrated layers may be performed using techniques that render the boundary between the layers imperceptible.
One method of performing this is by shooting a reference image such as a ramp or test image and using this image to achieve calibration in the transition area between channels. When multiple output channels are used, there may be a considerable overlap where both channels can legitimately contribute to the image. In this case, calibration is simplified because there is a region where weighted averaging between channels is applied.
A method for handling merging between channels is to define a function F(C1,C2) where C1 and C2 are signals from each of the channels and F defines the combined response between the two channels. This function is defined for the area of overlap or joining between channels. In general, C1 and C2 will be the calibrated response of each channel. Function F has coefficients M1, M2. M. V is the response of the combined channels as a result of the function F in the region of overlap and C1 and C2 in the region outside overlap and
Filtering the function V produces a smoothed version of V, Vs. By computing using Newton, Conjugate, or related techniques, the minimal error defined is the sum of the pixel errors across a range of images from the sensor, where the pixel errors are defined, for each pixel, as
Optimization selects the values of the coefficients M, to produce an optimal function, F that will be determined under the appropriate constraints. The nature of these constraints depends on the specific data being created and specific form of the functions being implemented. It is often necessary to experiment with different constraints before the optimal solution is found. Constraints generally are associated with bounding the available ranges of the coefficient M. Thus, constraints may be experimentally derived, for example, to constrain the values of M to converge on a solution or due to constraints that arise from the physical limitations of the software or hardware implementing the function F. In some cases, although data C1 and C2 are calibrated, the function F will be a function of individual pixels. This can occur when the point at which channel overlap occurs varies pixel-by-pixel. In this case a number of solutions are possible as previously discussed. Once the function V has been computed, it can be imbedded in a lookup table as shown in
When calibrated, data normalized to the switch point is used, as shown in
Sensors designed for accommodating highlights, or highlight sensitive sensors can be provided in a number of different ways as previously discussed. In another embodiment, the maximum CCD well voltage is set to a certain value for most of the integration time. Then, the well value is briefly increased.
However, a cell with a high intensity of light falling on it is shown increasing its voltage along line 2388 and reaches saturation at point 2390. It then maintains saturation level 2392 during the remainder of integration period T1. Then, waveform 2386 is increased during time T2, and because the brief period of the integration time T2, although the voltage increment is small between saturation levels 2392 and 2394, as indicated by voltage V. The site does not reach saturation during time T2, and therefore is able to be discharged during time T3, producing a voltage that is related to the intensity of the highlight. Calibration of this system is performed by measuring cell response described herein thereby accurately determining saturation level 2394 at the output of the electronic sensor.
Alternatively, in another procedure embodiment, the threshold level may be set for each photosite by determining the saturation level 2392 at the output of the electronic sensor. This may be done during the exposure to highlights in a procedure that has no second period T2. This procedure ensures that all cells are charged to the saturation level. This enables the calibration process to measure the sensor output corresponding to the saturation level 2392 for every photosensitive site in the sensor.
The saturation level 2392 may not be constant but may be influenced by the amount of light falling on the photocell due to resistance in the path between the photo site and the saturation reference voltage source. This resistance may occur at both the source and drain or cathode and anode side of the photo site. This resistance will only be apparent when the photo site is in the saturation mode as it is only at that time that current is flowing through the photo site into the reference voltage source. This resistance will also affect the apparent level of the saturation level 2392, the reference voltage. Thus, part of the process must be able to account for variations in this resistance from site to site. If the technique is not able to account for the variations in this resistance, the highlight accuracy will suffer. Once saturation level 2392 is determined, the slope of each mode can be determined, and these values can become parameters in the computation of the highlight calibration lookup tables referred to in the herein.
Specific sources of variation from photosite to photosite (and also with regard to temperature and light intensity) include: photosite dark current, photosite sensitivity, photosite linearity, photosite series resistance 2404, photosite series resistance 2406, resistor 2416, current source 2402, series resistance 2412, voltage source 2408, variation in the T1 integration time, variation in the T2 integration time, and CCD discharge gate resistance.
By calibrating using the measurements and analysis procedures described herein these variables are implicitly addressed by measuring the response of the light sensor or sensors at a calibrated range of values. As a result, data is produced that takes into account all of these factors that have an impact on the sensor response to highlights and across the entire range of the sensor. Thus, the method for determining light intensity from the output of this particular device architecture is to use the calibration methods outlined herein. The ratio of T1 to T2, coupled with the voltage V and the sensitivity of the photo site, determines the amount of additional range that this technique provides. Other components that have some effect on this function are the values of series resistance 2404, 2406, and 2412, voltage sources 2408 and 2418, and resistor 2416.
If an approach with multiple break points is required, then the process explained in
In an additional embodiment, a further technique is to use a separate CCD or sensor device and image a highly attenuated version of the image onto the CCD using optical methods. This is shown in
It is also possible as shown in
Because specular highlights will inevitably result in overflow of the CCD well in areas of the image where the highlight occurs, another method is to ensure that this overflow generates small calibrated increases of well voltage which can be later processed to compute the value of the specular highlights. This may involve using a diode or similar structure used to produce the increases.
This is illustrated in
Alternatively, a CMOS element in a source follower or drain follower configuration may be used to enable the discharge in either
A further method is to use a single CCD site with multiple resets. As that site reaches close to the saturation value within a certain fraction of the integration time, the specific site is reset. In the remaining time, a faster integration is performed, thereby rendering the CCD site less sensitive.
The timing configuration for this approach is illustrated in
The advantage of this approach is that both charging cycles begin at the same voltage level. However, it is still necessary to have a calibration technique that can accurately calibrate in spite of changes of the level 3104, so that a seamless image may be achieved for areas where photo site charge level would be close to 3104 at time 3103—that is within the region where this process transitions from the single charge mode to the dual charge mode.
There are a variety of ways to communicate to the external control electronics and calibration electronics that must process the resulting signal. One way is to actually discharge through the CCD transfer mechanism the output signal at time 3105 and then again at time 3107. In the case of
An alternative approach to signaling which cycle the site discharge corresponds to is to exclude charge ranges that occur during the output of the single discharge cycle (
Another approach is to use multiple discrete exposures with different integration times such that the resulting image dynamic range covers the entire range of the scene. This multiple approach may require motion compensation between frames. This is illustrated in
Photographic film is often able to capture the full dynamic range of a scene. However, this full dynamic range is generally not communicated to the viewer through conventional optical techniques. However, it is possible to scan the photograph using digital scanning techniques, and capture the full dynamic range that is found on the film. Thus, it is possible to produce a digital signal that represents the dynamic range. Digital intermediate is a term used to describe the process of inserting a digital image processing after film origination or before film output.
Another term that is used to describe this approach is hybrid imaging. The term is used because the approach as illustrated in
An example of this technology is the Cineon scanner using Cineon color space image data described in SMPTE standard 244M. In highlight regions, this is not specifically calibrated unless the techniques disclosed herein are employed.
The present invention has been described with respect to multi-mode electronic image sensors. However, the present invention can also be applied to scanned film where highlight corrections/shaping is possible by identifying and correcting highlight regions.
The many features and advantages of the invention are apparent from the detailed specification and, thus, it is intended by the appended claims to cover all such features and advantages of the invention that fall within the true spirit and scope of the invention. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the invention to the exact construction and operation illustrated and described, and accordingly all suitable modifications and equivalents may be resorted to, falling within the scope of the invention.