Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
  • H04N1/40—Picture signal circuits
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
  • H04N1/00795—Reading arrangements
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
  • H04N1/04—Scanning arrangements, i.e. arrangements for the displacement of active reading or reproducing elements relative to the original or reproducing medium, or vice versa
  • H04N1/12—Scanning arrangements, i.e. arrangements for the displacement of active reading or reproducing elements relative to the original or reproducing medium, or vice versa using the sheet-feed movement or the medium-advance or the drum-rotation movement as the slow scanning component, e.g. arrangements for the main-scanning
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N2201/00—Indexing scheme relating to scanning, transmission or reproduction of documents or the like, and to details thereof
  • H04N2201/04—Scanning arrangements
  • H04N2201/0402—Arrangements not specific to a particular one of the scanning methods covered by groups H04N1/04 - H04N1/207
  • H04N2201/0404—Scanning transparent media, e.g. photographic film
  • H04N2201/0408—Scanning film strips or rolls

Definitions

• regression data is accumulated during development to describe a curve of density versus time of development for each pixel. After development, this regression data is used to recreate a regression curve of density versus development time for each pixel. The time at which this curve crosses a density known to give optimum grain characteristics, called the optimum density curve, is used to create the brightness for that pixel in the finished stitched image.
• the invention further teaches weighting regression data as a function of time and density generally following proximity to the optimum density curve.
• a scene 102 portrayed as perceived through the wide dynamic range of the human eye, has highlights 104, midtones 106, and shadows 108, with details in all areas.
• a camera 110 is used to project the scene onto a film inside the camera.
• the scene is perceived by the film to consist of points of light, each with an exposure value which may be mapped along an exposure axis 112.
• the film is removed from the camera after exposure and placed in a developer.
• an electronic camera 120 views the film by nonactinic infrared-light during development.
• the film 122 still has a low density for shadows 124 and midtones 126, but may optimally reveal highlights 128.
• the infrared camera 120 inverting for the negative of conventional film, the shadows and midtones 130 appear black, while the highlights 132 are seen more clearly than at any later time in development.
• Clarity may be defined technically as the best signal to noise ratio, where signal is the incremental change in density with exposure, and noise is the RMS deviation in density across a region that has received uniform exposure, by convention scanned with a 24 micron aperture. For example -3-
• the midtones 126 typically have too low a density, or are too dark, to have enough of a signal level to reveal detail through the noise of the film and capture system.
• the midtones 164 are "washed out", such that not only is their contrast, or image signal strength, too low, but the graininess of an overdeveloped silver halide emulsion gives a high noise.
• the midtones 140 have developed to an optimum density that yields the best signal to noise ratio, or image clarity, for that particular exposure value.
• the shadows reach optimum clarity at four minutes of development 160, and the highlights reach optimum clarity at one minute of development 128.
• the optimum density will be different for different exposures, as in this example wherein the shadows 160 reveal best clarity at a lower density than the highlights 128.
• the section of the density curve around the optimally developed shadows 160 is copied as segment 170.
• the density curve around the optimally developed midtones 140 is raised on a base value, or pedestal 172, and copied next to curve 170 as curve 174.
• the height of the pedestal 172 is adjusted so the two curves 170 and 174 align.
• the curve around the optimally developed highlights 128 is adjusted and raised on pedestal 176 to produce curve 178. .4.
• the primary object of the invention is to merge images of differing densities into a single image which is free from the artifacts encountered in the prior art.
• a related object is to merge images of differing densities free of edge contouring. -5-
• a further object is to merge images of differing densities with reduced effect from nonimage noise.
• a further object is to merge images of differing densities while compensating for a shift in a density-affecting parameter, such as time. Another object is to recover missed images in a series of images of differing densities that are to be merged.
• a series of images are captured electronically from a developing film, each tagged with the time of capture.
• regression parameters are calculated, such as density times time, or density times time squared. These parameters for each time are summed into parameter accumulating arrays.
• the parameters can be weighted prior to summing by a factor sensitive to the reliability of each sample.
• the regression statistics are not necessarily viewable images, rather they describe in abstract mathematical terms smooth continuous lines for each pixel that pass through the actual sampled densities for each pixel.
• Figure 1 depicts the prior art of electronic film development with stitching.
• Figure 2 further portrays prior art stitching.
• Figure 3 portrays density versus time of a typical development cycle.
• Figure 4 introduces the graphic basis of the invention.
• Figure 5 adds to Figure 4 the effect of measurement noise and timebase shift.
• Figure 7 portrays an unweighted embodiment of the invention schematically.
• Figure 8 illustrates a problem with an unweighted embodiment.
• Figure 9 portrays weighting proportional to proximity to an optimum density curve.
• Figure 10 presents the preferred embodiment as a series of steps.
• Figure 11 presents the preferred embodiment schematically.
• the optimum density point for highlights 328, for midtones 340, and for shadows 360 lie on a locus of points called the optimum density curve 370, shown by a dotted line in Figure 3. This curve is found empirically by measuring signal to noise ratio for varying exposures, and finding the density at which each reveals detail with the optimum clarity.
• An image can be thought of as consisting of an array of points, or pixels, each of which receive a specific exposure.
• the prior art of Figure 1 thought of this exposure as-producing a specific density.
• each development time produced images of differing density that could be cut, aligned in density, and merged together.
• Figure 4 suggests thinking of the image as consisting of an array of pixels, each having received a specific exposure resulting in a specific development curve, such as the highlight curve 402.
• Each specific development curve can be quantified by the time at which the density of the development curve crosses the optimum density curve 404. In this case a specific highlight pixel produces curve 402, which crosses curve 404 at point 406, which can be tagged as a one minute pixel.
• Figure 5 illustrates a more typical case wherein the electronic camera adds noise to the captured images.
• each capture will not only lie along a line that is a function of film exposure and grain, but in addition each capture adds a random noise deviation from the line.
• the individual density samples for a particular pixel are illustrated in Figure 5 as solid dots, such as dot 502 measured at a time of 2.3 minutes.
• the density of sample 502 differs from a theorized true curve 504 because of noise in the electronic camera arising from, typically, statistical errors in counting photons, called shot noise.
• a best fit curve 504 can be estimated, and the time this best fit curve 504 crosses the optimum density curve 506 calculated, as before.
• Well known statistic practices give an array of choices.
• a particularly interesting best fit function is the locus of lines representing the developed density versus time as a function of exposure level for similar film, such as curves 510, 512, 514 and 516.
• the curve with the best fit to the sampled data, curve 504 in this example, is selected, and the crossover time read directly as the variable.
• the best fit curve could also be described mathematically as F(t,e), where e is the exposure level yielding a particular curve.
• This locus of lines can be derived by actually developing a series of test films given known exposures, and storing the actual measured densities of each as a function of development time t in a lookup table where one axis of the lookup table is the known exposure, a second axis is the time since developer induction that a specific measurement is made, and the value stored in the lookup table is the empirically measured density.
• a curve repertoire from a repertoire of mathematically simple curves, such as a series of lines in linear regression or curves described by a quadratic formula in a quadratic regression, such a regression will be called an empirical curve regression after the locus of curves derived from empirically measuring film during development.
• Any specific curve can be read from the lookup table by selecting a particular exposure value as one axis of the lookup table, and then varying time, the other axis, while reading out density.
• the value of exposure yielding the curve with the best fit is the best fit curve.
• the parameters gathered to specify one of the repertoire of available curves in the lookup table can be derived by summing the density of each sample point times functions of time.
• a first parameter is the product of a first function of time
• a second parameter is a second function of time, and so forth.
• one of the functions of time is time squared.
• the first function can be related to an average of the curves.
• the second function can be the primary mode by which the curves differ from this average, called the residue after the first function has been subtracted from each curve.
• the third function can be the main remaining residue after removing the components of the first and second functions, and so forth to whatever order is desired.
• Another interesting parameter set is based on a gaussian function of time, where the gaussian function is taken to be a function of time that rises and falls smoothly in a bell shape to select a particular period of time.
• a series of parameters based on such overlapping gaussian functions would specify the smooth shape of a curve.
• the use of such curves may be found in the art, especially in spatial transforms where the human retinal neural system is found to respond to gaussian and difference of gaussian (DOG) functions.
• Figure 6 illustrates the case of a real time operating system that terminated capture prior to reading all the data.
• data as represented by solid dots
• the best fit method allows a curve to be found and projected through the optimum density curve. Although this would not produce the most grain free image, it would produce an acceptable image under conditions in which the prior art would have struggled because there were no middle and late exposures to stitch. This ability to recover from a failure is again a significant advance over the known art.
• Figure 7 inputs the same sequence of seven scanned images 702 and requires the intermediate storage of three accumulating arrays 704, 706, and 708, as shown in the prior art example of Figure 2.
• the distinction over the prior art is that in Figure 7, the accumulating arrays sum regression statistics rather than images. This is emphasized by portraying the three accumulating arrays 704, 706 and 708 with crosshatching to indicate they are not meant necessarily as viewable images, but rather as statistical data.
• a series of images 702 is received sequentially from an electronic camera viewing the developing film.
• image 720 is received at two minutes of development.
• the density of each pixel of image 720 is summed with the density of corresponding pixels of images taken at other development times, such as image 722 at five minutes.
• This summation occurs in function block 724 which can either operate on all images together if they have all been accumulated and stored in memory during development, or one by one as they are captured from the electronic camera at the corresponding development times.
• the advantage of summing them as they are captured is that less memory is required.
• the resulting sum from function block 724 is stored in the accumulating array 704.
• the density of each pixel of image 720 is multiplied by time, two minutes in this example, and the product summed with the density times -10-
• a best fit curve is derived for each pixel by retrieving the corresponding statistical data for that pixel from the accumulating arrays 704, 706 and 708.
• the time of intersection of the best fit curve for each pixel with the optimum density curve is calculated in function block 740, and a function of this time stored for the corresponding pixel in the finished image array 742.
• the function stored in the final image array 742 can be the time directly; or it can be the exposure known empirically to develop to the particular time, found empirically and expressed in the computer as a function, such as through a lookup table, of the time; or it can be any other function found to-have utility, such as the square root of the linear exposure value normalized and stretched to fit white level and black level, yielding a conventional 8 bit computer image representation.
• a specific highlight exposure curve 802 is plotted as density versus development time.
• a series of noisy samples are represented by solid dots such as dot 804 representing a noisy measurement of density at one minute, and dot 806, representing a noisy measurement of density at five minutes.
• the goal is to find the crossover time 808, one minute in this example, where curve 802 crosses the optimum density curve 810, with the added constraint that this point 808 be found without exactly knowing curve 802, only the noisy measurements such as 804 and 806.
• weighting factors are assigned such that a linear regression yields straight line 912, which is much closer to the true curve 902 in the region around the crossover time 922, at the expense of deviating at outlying times, such as six minutes, for which accuracy is irrelevant.
• one object of this invention to reduce noise is accomplished via application of a weighting factor.
• the weighting function should generally follow the reliability, or signal to noise ratio, of each sample.
• the crest of the weighting function should therefore follow the optimum density curve, shown in Figure 9 as the 100% weighting curve 930, and fall off in proportion to distance from this curve.
• sample points such as 904 that are close to the optimum density curve are shown large, while sample points such as 906 that are distant are drawn faintly.
• the falloff would be a continuous function of distance.
• the rate of falloff with distance follows the rate of falloff of overall signal to noise ratio including both film grain and noise in the electronic capture system, and is found empirically.
• Figure 9 further portrays the true curve 940 for a shadow exposure with the actual sample points from which the true curve is to be mathematically estimated.
• sample points such as 942 that are proximate to the optimum density curve 930 are portrayed large to indicate a high weighting, while outlying sample points, such as 944, are drawn faintly.
• a best fit linear line through the large points around curve 940 will closely approximate curve 940 around five minutes at which time the curve crosses the optimum density -12-
• Curve 940 is included to point out that a weighting factor proportional to distance from the optimum density curve will tend to emphasize samples at short development times with highlight exposures, and samples with long development times for shadow exposures.
• the weighting factor should not drop to zero at very low densities or short times, but should maintain some finite weight. This insures that even if the system fails to capture samples at later times, there will be some regression data in the accumulating arrays from which a reasonable guess at the best fit curve can be extracted. Therefore, sample point 944 would be given a small but non-zero weight. The weight can, however, go to zero for high densities and long times.
• Figure 10 portrays the embodiment just described as a series of steps such as would be implemented in a computer program. The specific example given is one of linear regression.
• a first function receives a density and time as parameters and returns a weight factor
• a second function receives time as a variable and returns an optimum density
• a third function receives a crossover time as a parameter and returns a brightness value artistically representative of the light the crossover time represents, which is typically the square root of linear brightness, called a gamma correction.
• an electronic camera views the film.
• the image from the camera is digitized into pixels.
• each point from the film is sensed at several times.
• the process receives these pixels one at a time, receiving for each pixel four numbers: a density, x and y coordinates of the pixel, and a time at which the specific density was measured.
• regression parameters are calculated and collected. Specifically, regression parameters are calculated based on density and time, and array elements in the regression summation arrays pointed to by x and y are incremented by the regression parameters.
• the summation arrays are complete, and the data they hold can be used.
• regression data is read from the arrays and used to calculate a linear best fit curve.
• the time at which this best fit curve crosses the optimum density curve is calculated, or read from a lookup table, or solved by iteration.
• the equivalent x,y element of a final image is set according to a gamma function of this crossover time.
• Figure 11 covers the procedure of Figure 10 schematically.
• a series of images are received from an electronic camera viewing a film at specific times since a development has been induced.
• image 1102 is received for two minutes of development.
• a function such as 1104 labeled "weight” receives the time and also receives the numeric density of each pixel in the image, returning for each pixel a first value that sums to the corresponding pixel in a "summation weight" array 1106 via conduit 1108 terminating in 1110. This first value is also sent to multiplication block 1112 along with the numeric density of each pixel to produce an output product "summation density" for each pixel.
• “Summation density” sums to the corresponding pixel in "summation density” array 1114 via conduit 1116 and 1118.
• Another multiplication block 1120 also receives this "summation density” for each pixel, further receives the corresponding time, and outputs a product called “summation time” for each pixel that is the product of weight, density, and time.
• “Summation time” sums to the corresponding pixel in the "summation time” array 1122 via conduit 1124 and 1126. The process is repeated for each new image, such as image 1128, that is received by the electronic camera captured at a different development time. Following development, the three parametric arrays 1106, 1114, and 1122 are used to calculate a brightness value for each pixel.
• a faster method divides the two values from the summation weight array and summation time array by the corresponding value from the summation density array, and the resulting two numerical values are used as a pointer into a two dimensional lookup table that holds the precalculated time, or a gamma function of the time.

Landscapes

• Engineering & Computer Science (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Image Processing (AREA)

Applications Claiming Priority (3)

Publications (1)

Family

ID=22126569

Family Applications (1)

Country Status (4)

Families Citing this family (4)

Family Cites Families (3)

• 1999
  • 1999-02-22 AU AU33069/99A patent/AU3306999A/en not_active Abandoned
  • 1999-02-22 EP EP99934385A patent/EP1078505A1/de not_active Withdrawn
  • 1999-02-22 WO PCT/US1999/003845 patent/WO1999043148A1/en not_active Application Discontinuation
  • 1999-02-22 TW TW88102561A patent/TW384416B/zh not_active IP Right Cessation

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events