CA1283477C - Correction of shading effects in video images - Google Patents

Abstract

CORRECTION OF SHADING EFFECTS IN VIDEO IMAGES

Abstract of the Disclosure Video images are corrected for shading effects by generating a histogram of picture element (pel) gray scale intensity values and calculating the median and the black and white extremes thereof. A range of permitted background gray scale values is empirically selected. A background image is then created by sampling the intensity values of the original image pels and using a look-up table to remap the image. In one embodiment, pel intensity values are compared with the median value and only corrections lying within the range are carried out. A duplicate image is created from the remapped image and the intensity values of a number of pels in a predetermined region about a first pel in this image are sampled and the value of the corresponding pel in the background image is changed to the largest value among those of the first pel and the sampled pels. This is continued using a look-up table with a 7 bit output until all of the pels in the background image have been changed accordingly. The duplicate image is then changed to conform to the changed background image and the sampling procedure is carried out over a larger surrounding region.

Description

~;~8347'7 CORRECTION OF SHADING EFFECTS IN VIDEO IMAGES

DESCRIPTION

BACKGROUND OF THE INVENTION

1. Field of the Invention The present invention relates to image data enhancement apparatus and methods, and more particularly to apparatus and methods of flltering undesired shading effects from video images.

2. Description of the Prior Art 1~83477 In the prior art there are many image data processing systems available. Some examples of systems representative of the prior art are the following.

U.S. patent 3,621,129 to Fisher relates to devices for enhancing images by quantizing picture element levels into binary digital signals employing a threshold which is derived from the gray scale values of the image under analysis. The apparatus of the patent examines gray scale values of picture elements immediately preceding and succeeding a feature under analysis and a variable threshold i~ generated which i~ derived from a mean value between the maximum black signal level immediately before and after the feature under analysis.

The patented system does not acquire a histogram of distribution of gray ~cale intensity levels in the entire image to provide a uniform correction but rather provides only a localized threshold calculation.

U.S. Patent 3,979,555 to Oppitek relates to a histogram equalized system that adaptively redistributes the intensity levels of video signals in order to provide an Yo9-82-040X 3 equal number of elements at all display intensity levels.

This patented system does not relate generally to a method of correcting shading effects in video images where a correction value is calculated from the variations in background level in the entire image.

U.S. patent 3,800,078 to Cochran teaches a digitally compensated scanning system wherein an initial scan of black or background information is stored in a look-up table and compared to desired data signals to eliminate undesired variations or noise resulting from photodiode leakage current or other noise sources.

This patent does not show a method of correction shading effects in video images in which a histogram of distribution of gray scale intensity levels i~ acquired and u~ed to calculate a threshold value and range of permissable background gray scale values that generates a correction signal which, when combined with the raw data provides an output signal that is corrected for undesired shading effects.

U.S. Patent 4,232,400 to Yamamoto shows an echo control system for an adaptive echo canceller which cancels an echo signal while successively estimating the transmission characteri~tic of an echo path that includes means for filtering by convolution a transmitted signal including echo characteristics.

The patent does not teach a method of correcting ~hading effects in video images using a histogram of distribution of gray scale intensity levels for calculating maximum white, maximum black and threshold values and generating a correction signal to eliminated shading effects.

An article entitled "Digital Processing Techniques for Encoding of Graphics" by Ting and Prasada, which appeared in the Proceedings of the IEEE, V01. 68, No. 7, July 1980 at page 757 and following, providec a survey of techniques for preproce~ing input documents having noise contents which croate poor data compression ratio~.

The article discusses a piece-wise local approach to shading correction in which a number of ~mall overlapping samples or windows are examined and a threshold is established for each of these windows. The approach presents a potential problem with respect to different thre~hold~ at either side of a boundary between adjacent windows. Discontinuities in the threshold going form window to window may result in undesired effects in the image. The article does not present a global shading correction algorithm in which the shading for the image is corrected without discontinuity.

Other articles digcussing the general problem of thresholding are found in IEEE Transactions on Systems, Man, and Cybernetics, Vol. SMC-8, No. 8, August 1978, at pps.
622-632. The segmentation of an image into background and foreground regions is discussed as well as the general technique of background sampling in the region of an object and the difficulties in the choice of background samples.

However, a slmple and effective method and means for correcting shading efects in video image~ i8 not found among this art.

It will therefore be seen that the prior art discussed above neither teaches nor suggests the present invention as disclosed and claimed herein.

lZ834'77 SUMMARY OF THE INVENTION

Accordingly, it is an object of the present invention to provide methods and means for correcting shading effects in video images having a plurality of picture elements where each picture element or pel may be represented by a gray scale intensity value. The method of the invention involves e.g., in suitable computer means, the steps of generating a histogram of distribution of gray scale intensity values of picture elements in the original image, and calculating a maximum white gray scale value, a maximum black gray scale value and an intermediate gray scale value, such as the median, from the histogram. The image polarity, that is the background colour. may then be determined empirically by noting which of the gray scale extremes is located closest to the intermediate value. A range of permitted background gray scale lnten~ity values i9 then selected such as by calculating the distance between the median and the background or ma~ority colour extreme and setting the range as twice that distance centered at the median. Having established these parameters, two approaches are presented for correcting for shading effects.

~83477 In the first embodiment a lookup table (LUT) iq constructed which remaps the original image to a correction image. The original intensity value of each pel i5 compared with the median value and the LUT output represents the difference between the two. If the difference is outside the permitted range, no correction is made; otherwise, the difference is used to correct the original intensity. A
neighbourhood sampling procedure is carried out for each pel to obtain an average value for each correction.

In the latter embodiment a first background image is created, e.g., in a "reference" buffer, by sampling the gray scale intensity value of each pel in the original image to be corrected and using a suitable lock-up table ~LUT) to remap the image from 8 bits to 5 bits with all values outside the permitted or acceptable range set to that of the neare~t colour extreme. A duplicate background image is then created from the remapped image, e.g., in another or "neighbouring" buffer. Sampling of the gray scale intensity values of a number of pels in a predetermined region around a first pel in the duplicate image is then carried out and the gray scale value of the corresponding pel in the first background image to be corrected for shading is changed to the largest . .

value (assuming the majority colour extreme to be larger) among those of the first pel and the sampled pels. This operation may be accomplished using a 10 bit LUT by making the 5 bits representing the first pel value the low order input and those of the sampled neighbouring pel the high order input. The duplicate background image is then changed to conform to the changed first background image and the sampling procedure is carried out over a larger surrounding region and a larger number of pels with the corresponding changes in the first background image. This procedure may be repeated until a background image with the desired level of shading corrections is achieved.

It i# another object of the present invention to correct shading effects in video images as above by further including the step of determining, during the sampling over the ~urrounding reglon, if the corrected gray scale value of oach pel in the background image is less than a predetermined value. This is accompli~hed by detecting "low" pel values or poor whites (poorwht) and storing indications of such in an overlay plane, which . j~ .

indications are changed when appropriate during successive samplings. subsequent samplings may then be carried out only on pels with low value indication8 until the desired correction level i 8 achieved.

It is yet another object of the present invention to correct shading effects about the edges of video images and a method and means for this purpose is also described.

Therefore, an apparatus for correction shading effects in video images according to the present invention employs an image proce~sing system including: a plurality of full image buffers; means for generation an histogram of, and providing the histogram calculations for, the image to be corrected; means for creating a background image and a duplicate background image in the buffers; means for controlling the operation of the image processor for sampling the gray scale inten~ity values of picture elements ln a region around a current picture element to be corrected in the duplicate background image; means for correcting the value of the corresponding current picture element in the background gray scale value determined from the previous step; means for executing the sampling and lX83477 correcting steps for each picture element in the duplicate and background images and conforming the former to the latter upon completing all of the correction; and means for repeating the proceeding ~tepg to provide a uniform shading correction throughout a video image.

The foregoing and other objects, features and advantages of the invention will be apparent from the more particular deæcription of the preferred embodiments of the invention, a~ illuQtrated in the accompanying drawing.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1.1 illustrates a character of a video image intercepted by a scale line to produce a video signal.

FIG. 1.2 shows an idealized video signal generated from tho scannod character of FIG. 1.1 FIG. 1.3 show~ a video signal as in FIG. 1.2 including ehadlng defects from camera characteri~tics, lighting or the like.

~83477 FIG. 2 is a general block diagram of apparatus according to the present invention for correcting shading defects in a video image.

FIG. 3.1 is a diagram illustrating a histogram of the number of occurences of pel intensity values before shading corrections.

FIG. 3.2 is a histogram as in FIG. 3.1 after shading correction has been performed in accordance with the apparatus and method of the present invention.

FIG. 4.1 is a general diagram of the steps involved in analyzing a histogram to determine parameters to be used for the shading correction.

FIG. 4.2 i~ a flow chart of the steps used in doterminlng image background colour.

FIG. 4.3 is a flow chart of the steps used in determining the poor white (POORWHT) range.

FIG. 5.1 i3 a general flow chart illustrating the operation of the shading correction filter.

1~83477 FIG. 5.2 shows the steps involved in inverting the image when the background colour has been determined to be black.

FIG. 5.3 shows the steps involved in creating a background image in a five bit representation.

FIG. 5.4 shows the steps used to create the look-up tables for selecting the whiter of two background image values .

FIG. 5.5 is a flow chart of the initial maximum white selection process carried out in a small neighbourhood around each pel of the background image.

EIG 5.6 is a flow chart of the maximum white selection process carried out in large neighbourhood wherever the maximum white image remains in the POORWHT.

EIG. 5.7 is a flow chart of the summation process used to filter any noise in the maximum white background image.

EIG. 6.1 i8 a diagram showing an uncorrected video signal having a concave ~hading defect.

~83477 FIG. 6.2 is a diagram showing the concave shading defect isolated from the video signal in accordance with the present lnvention.

FIG. 6.3 is a diagram showing the shading corrected video ~ignal in accordance with the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Most images captured from television cameras have substantial shading problems, even when the cameras incorporate shading correction circuitry. The shading, or low a partial frequency variations, in the image can cause problem~ when the image is to be converted by thresholding to a graphics image having one one or two bits per pel (picture element). Apparatus and methods are disclosed herein for correcting these shading effects and while those skilled in the art may adapt the invention to use in many different ways drawing from the following description, the present invention may be particularly implemented using a commercially available image proce~sor, such as a Grinnell GMR-270, operating under the control of a digital data processing ~ystem, such '~.

~83477 as the IBM* Series I, so that the description will be directed to such an implementation.

Referring firstly to FIGS. 1.1, 1.2 and 1.3, the generation of a video signal with shading effects will be discussed. FIG. 1.1 shows a text data character such as 0 being intercepted by a horizontal scan line.

Ideally, a black character on a white paper background would produce a video signal such as is shown in FIG. 1.2 having a relatively uniform flat background white level except for areas where black text data occurs at which point the video signal would switch to the black level.

As can be seen in FIG. 1.3, the actual video signal representative of the 0 of FIG. 1.1 includes shading effects which create ambiguities in separating text data from background in an image. It can be seen in FIG. 1.3, that the shading of the white paper background causes a gray #cale variation in areas where no text data appears which can inhibit efficient data compression and transmission and create undesired artifacts in the transmitted image.

* Registered Trademark Referring now to FIG. 2, a block diagram of apparatus according to the present invention is illustrated. A video signal containing image data is stored in image buffer 12 in such a way that access may be had to an intensity value for each pel in an image. The data in image buffer 12 is ~ampled by an histogram acquisition circuit 14 to determine the gray scale level distribution of pels in the image. FIG.

3.1 shows the form of a resulting histogram for a graphics image before shading correction. Note that for a black data-on-white background graphics image the majority of pels are normally near the white intensity values since in a black-on-white graphics image there are usually many more background pel positions than there are data pel positions.
There are many techniques for acquiring and presenting histogram information including that shown in U.S. Patent 3,979,555 discussed above. Once the histogram has been acquired it is then analyzed to determine certain working parameters to be used in carrying out the shading correction. This analysis may be accomplished in a general purpose computer 16 and the output provided to a ~hading correction filter 20 which may be an image or other suitable processor.

1'~83477 FIG. 4.1 shows the sequence of steps to be followed in deriving parameters from the histogram for used in shading correction techniques described in this application.

The first three steps in FIG. 4.1 involve calculation of the median (MED), minimum (MIN) and maximum (MAX) gray scale intensity levels in the image. The calculation of these parameters and the calculation of the background colour may be carried out in various ways. The simplest technique would be to identify levels in the image. The calculation of these parameters and the calculation of the background colour may be carried out in various ways. The simplest technique would be to identify the background colour with the extreme closest median, and this will work well for most images. However, experimental studies have shown that unless an image background can be clearly classified as black, it should be cla~ified as white. This cla~slfication technique is shown in FIG. 4.2. For convenience, define WDIF a~ the difference between the white extreme (MAX). For an image background to be classified as black, 2 x BDIG should be less than WDIF.

After the background colour is determined, another parameter, POORWHT, is derived which is used as a measure 1~3477 of the range of background intensities for which the shading correction requires sampling over larger neighbourhood as will be described. The procedure for determining POORWHT, shown in FIG. 4.3, is derived empirically. Similarly to define an acceptable background range R, we define DIF as the magnitude of the difference between the median and the background extreme. The lower limit of the background intensities LN is set at MED-DIF, and the upper limit LP is MED + DIF. The total background range R is from LN to LP.
If the background intensity range can be cleanly discriminated from foreground intensities, LN is set to O
and LP is set at 31. This permits range R to be identified by 5 bits, as will be seen. POORWHT is then set to 2. All images classified as black background are automatically placed in this class. If the discrimination is not very good (WDIF2 2 x BDIF). POORWHT is increased to 5, and if the discrimination is very poor (MIN2 MED-DIF), POORWHT is increased to 12. The intensity range from MIN to MIN + R x (POORWHT-1)/32) ls defined as the background intensity range where the shading correction requires sampling over a larger neighbourhood .

Once the background colour has been identified and MED, DIF

and POORWHT obtained from the histogram, the shading 1'~83477 correction is performed. Two embodiments will be described but first it should be understood that the shading correction methods and apparatus according to the present invention are based upon the following sampling concept.
Gray scale values of the pels in a general neighbourhood around a given pel in an image to be corrected are sampled.
the samples must be far enough from the pel being corrected and from each other that any data related information content shows essentially no correlation between samples or with the pel being corrected. The samples should be closely enough spaced, however, that an accurate estimate of the low spatial frequency shading effects can be obtained. As a practical matter, samples should be taken at a radius distance from the given pel greater than the character size of ht text data and perhaps in the range of 10 to 20 pel distances from that pel.

In a first embodiment of the invention a look-up table is constructed whlch remaps the original image to a correction image. Access is made to a look-up table which has as an input a multibit digital data signal representing the gray scale intensity value for one of a number of samples in the selected neighbourhood around a given or current pel being corrected. The output from 1'~83477 this look-up is a signal representing the difference between the input intensity value of the neighbouring pel being sampled and the median previously calculated, except that if the difference value determined exceeds the permitted range R for background, as determined above, the corrective output of the look-up table for that current pel is set to ~ero, representing no correction made. In this embodiment of the invention, further samples are taken in the selected neighbourhood point by point around the current pel to be corrected and the corrective outputs or correction terms generated by the look-up table are summed. An average value is obtained by dividing the sum of the correction terms by the number of samples taken. This averaged correction value is used to correct the current pel intensity value for shading effects in shading correction filter 20. The shading corrected intensity value of the current pel is determined by the following equation:

X' = X + (M-A) where X' is the corrected value of the current pel;

X is the original value of the current pel;

1~83477 M is the median value; and A is the average value of the neighbourhood background determined above, 80 that (M-A) is the average correction value for the neighbourhood sample.

Thi~ creates a correction image which is the difference between the signal image and the median for all pixels in the originals image with intensities within the background image R. For all pixels in the original image at MIN or below in intensity, the correction image is zero.

Since the method set forth with respect to this embodiment determines an average background from a group of samples around the current pel, if samples are taken in an all black area such as off the edge of the frame. a skewed average will result causing an erroneou~ correction. Therefore, for pels in the band around the perimeter of the image, the average ls taken including sample~ at the opposite edge of the image as if the image had been wrapped around so that the corresponding opposite edges were ad;acent. Although ~his technique i9 an improvement since, for example, shading at the right 1~83477 and left edges of an image may not be the same due to external conditions.

Referring now to FIGS. 5.1 - 5.7, a second embodiment of the present invention which deals with the edge problem will now be described.

For purposes of explanation, a brief description of the Grinnell GMR-270 image processor will be given. The embodiment is not limited to his particular processor, however. The Grinnell GMR-270 image processor has four input look-up table (LUTs) each having an 8 bit input and 8 bit output. the eight input bits correspond to 256 different possible outputs. the 4 input LUTs are arranged in pairs to provide two 16 bit inputs to an arithmetic logic unit (ALU). The high order input tables provide the most significant 8 bits and the low order input tables provide the least significant 8 bits to their respective inputs. The output of the ALU is 16 bits and a carry bit. The 10 least significant bits of the ALU output can be fed to an output LUT which produces an 8 bit output. This output table therefore has 1024 input states, but only 256 unique output states. The ALU carry bit can be written to overlay bit planes. One of the overlays can be used to switch be-1;~8347~

tween two logical functions in the ALU. The other can be directed to the most significant unput bit of the output LUT
in place of the normal processor output to that input bit of the table.

Inputs to the image processor are obtained from several image buffers. In the description below two buffers will be used, on called the background image or reference (REF) buffer, and the other called the duplicate image or neighbourhood buffer. Panning circuitry in the GMR-270 can be used to provide an offset between an image in the reference buffer and an image in the neighbourhood buffer.
Note that an additional buffer is used to keep a copy of the original image, as the images in both the neighbourhood buffer and the reference buffer will be transformed during the development of the shading correction.

Referring again to FIGS. 5.1 - 5.7, the first step i9 to invert the image and parameter values if the background colour is black. The sequence of steps to do this is shown in FIG. 5.2. The "exclu~ive or' operation shown in FIG. 5.2 is equivalent to subtracting the value from 255. Inverting 1mages with blacX backgrounds is desirable for the implementation in th- Grinnell GMR-270, 1'~83477 since neighbourhood æamples which wouLd be taken from outside the image boundaries are forced to black in this implementation. Note that if the image is inverted before the shading correction is performed, it is reinverted after the shading correction in complete.

The six steps shown in FIg. 5.1 following the invert of the black background images are described in greater detail in FIGS. 5.3 through 5.7.

As is seen in FIG. 5.3, the upper and lower limits for background intensity, LP and LN are calculated. a typical gray scale signal may have 8 bits representing 256 discrete gray scale levels ranging from white to black. the high order input tables are cleared to give zero output for all inputs, and a table is written to a low order input table to convert the image from its original 8 bit form to a 5 bit background intensity image. The reduction can conceptually be divided into three steps. First, the intensity range of the image is restricted to fall within the limits LN and LP
by clamping all pel intensities above LP to LP and all pel intensities below LN to LN. Next, the minimum intensity is reduced to zero by subtracting LN from the result of the first step. Finally, the dynamic range of the resulting 1'~83477 image is scaled to the interval O to 31 (5 bits) by multiplying the intensity by 32/R. An example of a look-up table which will do this sequence as a single operation is shown in appended Table I. The reduction of the background intensity range to 5 bit representation is required only because of the signal path limitations of the particular image processor being used in the GMR-270.

The next step is the development of a background image which contains the maximum white value among a set of pels that includes the current pel being corrected and a sample of neighbourhood pels in a small region around the current pel.
FIG. 5.4 shows the steps required to set the LUTs for this operation. The first two steps in FIG 5.4 shows the steps required to set the LUTs for this operation. The first two steps in FIG. 5.4 set the LUTs for the neighbourhood background image input. Both tables have as input the 5 bit neighbourhood image. Their effect is to multiply the input image by 32. The high order LUT i~ shown in the top section of appended Table V. Note that the first POORWHT entries (from O to POORWHT-l) in the top section have the most significant bit of the output set. The third step in FIG.
5.4 sets the he LUTs for the reference image input. The high order table is zeroed except for the first POORWHT entries, where the table output is set to hex 80. The ALU will add the outputs of these two sets of input tables to create a composite 10 bit signal to feed to the output LUT. If both inputs are less that POORWHT, the addition of the two input signals in the ALU
will set the carry.

The fourth step in FIG. 5.4 creates the output LUT. The function of this table is to compare the two 5 bit pel values and choose the larger of the two. Table II
(appended) shows the look-up table transfer function for selecting a maximum white intensity value from the two inputs each of which have been mapped to five bit wide signals.

Once the LUTs have been set, the small neighbourhood maxwhite selection is done. FIG. 5.5 is a flow chart of the steps. This maxwhite selection technique allow~ tho sampling of a relatively large number of neighbourhood pels ln very few steps. The desired neighbouring pel offset values are broken into short lists, with three offsets per list. For each offset in a given list a comparison is made between the neighbourhood pel and the reference pel to select the maximum white. The comparison i8 done for all pels in the image and the ~X834'77 result is stored in the reference image buffer, replacing the existing reference image. When the list is complete, the reference image contains the maximum white intensity among the four values sampled, i.e., the reference value and the three neighbourhood values. The reference image is then copied to the neighbourhood buffer and, if more is to be done, a new list is started. At the end of the second list the maximum white has been obtained from a sample of sixteen pelæ. In practice only two lists are processed, but if a third list of three offsets were processed the maximum white would be obtained from a sample of 64 pels. Note that after each compari~on if both neighbourhood and reference background images re in the POORWHT range the overlay bit i9 set.

A flow chart for the large neighbourhood maxwhite æelection is shown in FIG. 5.6. The LUTs are the same and the basic selection process is identical to the small neighbourhood maxwhite selection, except that the offsets used are larger and only reference image pels where the bit in the overlay is set to flag a POORWHT background intensity are compared with the neighbourhood pels. If after processing a given oset, an acceptable maximum ~ .

1~834~7 white value is obtained, the bit is cleared in the overlay and the correction then remains unchanged for the reæt of the large neighbourhood offset values. If an acceptable maximum white value is not obtained, the process is iteratively repeated until a predetermined limit of iterations is achieved or an acceptable maximum white value is found.

At the end of the large neighbourhood maxwhite selection the maximum white estimate for each pel is the result of comparison of up to eighty different pels - the original reference pel and up to 79 neighbours in the vicinity of the difference pel. The comparison i8 obtained, however, with a relatively small number of processing steps. Note that for pels near the edges and corners of the image th e sample size i8 ~maller. All samples taken outside the image are black, and therefore are re~ected by the maximum white comparison process. However, since the sample size is still large, the probability of a bad correction value near edges and corners is extremely small.

After the maximum white e~timate is obtained, a filtering is achieved by summing three nearest neighbour intensity values with the current pel. The summation is replaced by a multiply by four at the edges of the image in accordance with the overlay control described above. FIG.
5.7 contains the flow chart for this filtering operation.

In operation, a mask is set in two separate overlay planes wherein the mask applies to the entire image is set to the "off" condition and to the band of pels around the perimeter of the image if set to the "on" condition. During normal processing of pels of pels in the centre of the image, the first overlay plane is set "off", causing the proce~sor to sum the reference pel with the neighbourhood pel. For pels in the edge band around the perimeter, the first overlay is set "on", inhibiting summation and leaving the reference pel intact for the first two summations over all pels in the image.

The second overlay plane controls the output LUT operation.
When the third and final summation is made, the second overlay is fod to the most significant bit of the processor output LUT. If the overlay bit i8 set "off", a protrait of the output LUT is ~elected which transfers the ALU output without change to the reference buffer. If the overlay bit is set "on", the multiply by four table shown in (appended) Table III is selected, and the input is multiplied by four before storage in the reference buffer. This technique greatly reduces the effect of edges on this filtering operation.

The result of the steps described in FIG. 5.7 is a smooth background intensity image having values from 0 to 127. The summation process averages any noise produced by the maximum white selection process.

As shown in FIG. 5.1, the background intensity image must now be remapped to a correction value, using a table which rescales the background image to the correct dynamic range and shifts the zero correction point to correspond to the median. Table IV (appended) gives an example of the low order and high order input LUTs required for this remapping.
The background image as interpreted by this table, is added to the original image. The result, clamped to the range from 0 to 255, i8 the Rhading corrected image.

The nonlinear shading filter described above provides a variable pass band filtering by enabling lower pass band whenever the correction derived from earlier operation indicates a failure of the correction together with ~'~83477 noise averaging as a final step with lock out of the averaging at the borders of the image.

It will be seen therefore that apparatus and methods have been presented for simply and effectively correcting for shading effects in video images particularly in graphics form.

For convenience, a specific example of basic code that can be used with the disclosed apparatus for histogram analysis and shading correction is appended.

Y09-82-040X lZ834~7 TABLE I

0202 0304 0505 0607 0808 O90a ObOb OcOd OeOe OflO 1112 1213 1415 1516 1718 1819 lalb lblc ldle lelf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lflf lffl lffl lflf lflf lflf lflf lZ83477 TABLE II
0001 0203 0405 0607 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0101 0203 0405 0607 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0202 0203 0405 0607 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0303 0303 0405 0607 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0404 0404 0405 0607 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0505 0505 0505 0607 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0606 0606 0606 0607 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0707 0707 0707 0707 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0808 0808 0808 0808 0809 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf 0909 0909 0909 0909 0909 OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf OaOa OaOa OaOa OaOa OaOa OaOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf ObOb ObOb ObOb ObOb ObOb ObOb OcOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf OcOc OcOc OcOc OcOc OcOc OcOc OcOc OeOf 1011 1213 1415 1617 1819 lalb lcld lelf OdOd OdOd OdOd OdOd OdOd OdOd OdOd OeOf 1011 1213 1415 1617 1819 lalb lcld lelf OeOe OeOe OeOe OeOe OeOe OeOe OeOe OeOf 1011 1213 1415 1617 1819 lalb lcld lelf OfOf 00f OfOf OfOf OfO OfOf OfOf OfOf 1011 1213 1415 1617 1819 lalb lcld lelf "~, , Y09-82-040X 1~83477 TABLE II (continued) 0111 1213 1415 1617 1819 lalb lcld lelf 1111 1213 141S 1617 1819 lalb lcld lelf 1212 1213 1415 1617 1819 lalb lcld lelf 1313 1313 1415 1617 1819 lalb lcld lelf 1414 1414 1415 1617 1819 lalb lcld lelf 1515 1515 1515 1617 1819 lalb lcld lelf 1616 1616 1616 1617 1819 lalb lcld lelf 1717 1717 1717 1717 1819 lalb lcld lelf 1818 1818 1818 1818 1819 lalb lcld lelf 1919 1919 1919 1919 1919 lalb lcld lelf lala lala lala lala lala lala lala lala lala lala lala lala lala lblc lcld lelf 'blb lblb lblb lblb lblb lblb lblb lblb lblb lblb lblb lblb lblb lblb lcld lelf lclc lclc lclc lclc lclc lclc lclc lclc lclc lclc lclc lclc lclc lclc lcld lelf ldld ldld ldld ldld ldld ldld ldld ldld ldld ldld ldld ldld ldld ldld ldld lelf lele lele lele lele lele lele lele lele lele lele lele lele lele lele lele lelf lfl 11f 11 11 11f lflf 11f lflf lfl 11 11 11f lfl lfl lflf lflf 3~

~83477 0004 080c 1014 181c 2024 282c 3034 383c 4044 484c 5054 585c 6064 686c 7074 787c 8034 888c 9094 989c aOa4 a8ac bOb4 b8bc cOc4 c8cc dOd4 d8dc eOe4 e8ec fOf4 f8fc 0104 O90c 1114 l91c 2124 292c 3134 393c 4144 494c 5154 595c 6164 696c 7174 797c 8184 898c 9194 999c ala4 a9ac blb4 b9bc clc4 c9cc dld4 d9dc ele4 e9ec flf4 f9fc 0204 OaOc 1214 lalc 2224 2a2c 3234 3a3c 4244 4a4c 5254 5a5c 6204 6a6c 7274 7a7c 8234 8a8c 9294 9a9c a2a4 aaac b2b4 babc c2c4 cacc d2d4 dadc e2e4 eaec f2f4 fafc 0304 ObOc 1314 lblc 2324 2b2c 3334 3b3c 4344 4b4c 5354 5b5c 6364 5b6c 7374 7b7c 8384 8b8c 9394 9b9c a3a4 abac b3b4 bbbc c3c4 cbcc d3d4 dbdc e3e4 ebec f3f4 fbfc 3~
~r Y09-82-040X lZ83477 TABLE IV

1414 1413 1313 1212 1211 1111 1010 lOOf OfO OeOc OeOd OdOd OcOc OcOb ObOb OaOa OaO9 0909 0808 0807 0707 0606 0605 0505 ffff fffe fefe fdfd fdfc fcfc fbfb fbfa fafa 9f9 f9f8 f8f8 f7f7 f7f6 f6f6 f5f5 f5f4 f4f4 f3f3 f3f2 f2f2 flfl flfO fOfO
efef efee eeee eded edec eccc ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb ebeb Sign Bit for Remapping to True Correction OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfO OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf OfOf ~283477 TABLE V

0404 040g 0404 0404 0505 0505 0505 0505 OaOa OaOa OaOa OaOa ObOb ObOb ObOb ObOb OcOc OcOc OcOc OcOc OdOd OdOd OdOd OdOd OeOe OeOe OeOe OeOe OfOf OfOf OfOf OfOf lala lala lala lala lblb lblb lblb lblb lclc lclc lclc lclc ldld ldld ldld ldld lele lele lele lele lflf lflf lflf lflf 0020 4060 808a cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0002 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO
0020 4060 80aO cOeO 0020 4060 80aO cOeO

1;~83477 Preferred embodiment of histogram analysis and shading perception filter pigmentstion for a GMR-270 function hstgr;
(process histogram of image to derive nalues for the maximum, minimum and median. Determine the background colour of the image and the range of the background colour of the image and the range of background values which would probably be considered back in the sampling process for the shading correcion. Histogram data is stored in cmdbuf. A 480x512 pixwl image requires 18 bit precision for the histogram data. To avoid edge effects, the histogram is obtained for a 472xS04 pixwl area centered in the middle of hte 480x512 image. The high order 16 bits of the histo-gram data re in the 1st 256 half words, hte low order 16 bits are in the next 256 half words. A half word is defined to be 16 bits. All varuables referenced in this code are half words in storage or 16 bit registers.) half hgh(base bl) hgl(base bl+512);
(First calculate the median. The 1st 256 words contain the high order 16 bits. Sum these, ignoring loss of precision of the least significant 2 bits) nr=472 (height of data for histogram) nc=504 (width of data for histogram) tvarO=srl nr 1 (number or rows in image/2) b2=expl srl nc 2 (number of columns in image/4) b2=-expl * tvarO expl b2 (b2 is nc*nr/8) bl=addr omdbuf (set to bottom of historgram data) bsubl=bl (save bottom) bl-+bl 510 (point to top of histogram data) bO-O (zero sum) be8in (sum over high from top down) bO=+ bO expl hgh if b2 lgt bO
bl=- bl 2 repeat endif endbe8in if bl llt bsubl bl=bsubl endif med=expl srl expl - bl expl bsubl bl=addr obdbuf bliml=bl 09-82-040X 1~834~
(zero total state cnt) bO=O (use for low order part of sum) (get min level - sum from bottom up) begin bO--~ bO expl hgh (sum high) if bO llt 32 bl=~ bl 2 repeat endif endbegin (now count down til 1st zero found) begin if hgl>O
if bl lgt bliml bl=- bl 2 repeat endif endif endbegin min= srl expl - bl expl bliml 1 ~now get max level - sum from top down) bliml=~ bliml 510 (point to top of histogram data) bl=bliml bO=O (low order word of sum) begin bO=~ bO expl hgh (sum high) if bO llt 32 bl=- bl 2 ~repeat endif endbegin (now go up til 1st zero encountered) begin if hgl>O
if bl llt bliml if bl=~ bl 2 repeat entif endif endbegin max=expl srl expl - bl - expl bliml 510 1 dif=- med min (black dif) lf sll dif 1 ~ - max med (if image is clearly white on black bkgnd) color=O (white on black background) else (image was prob black on white background) (black dif must be twice white dif for clean graphics images) color=l dif=-max med (white dif) endif poorwht=2 . (init for clean graphics image) lf color=l (white background) dif=-max med (recalc in ca8e grayscale image) '' 3~7 ~, . .

lX83477 if -med min ~ sll dif 1 (check for low dynamic range/gray scale) poorwht=7(low pass shading correction more often) endif if min > -med dif (really low dynamic range/gray scale) poorwht=12(do low pass still more often) endif endif endfn function deshade;
(Note: This implementation of the shading correction algorithm is explicitly for the grinnell GMR-270 image processor. This image procossor consists of a 16 bit ALU with two 16 bit input paths.
Data from image memorles is fed through four 8x8 lookup tables (tables O and 1 feed one input lf the ~LU, and tables 2 and 3 feed the other). The low order 10 bits of the output can be fed through a lOx8 output table (table 4) if desired. The processor also allows one to select between two different ALU funcions based on a select bit from a one bit/pixel memory called overlay 8. Either the most significant bit or the carry bit of the processor can be written to overlay 8. For processor operations 3 image memories, memO, mem6 and mem/, can be used. MemO can be panned to address neighbouring plxels.
Correct for shading of the video imsge as follows:
1. Image is in memO,6 and 7 at start. All processor tables re 1:1, but overlay B.8 i8 still set. If color=O, invert in 0.7. Write look up talbe for med-video dif (5 bit output) with min white=O, msx white=31, allowable white range med+/-dif.
Remsp im~ge to 5 bit dlf in 0.6. Reset talbes. Wrlte table4 to select max white between two pixels.
2. Do neighbourhood convolution to select mixwh te for block 4 pixels.
Write overlay 8 wherever maxwhite values are both small. Final step of convolution wlll leave overlay 8 set where maxwhite selec-tion did not work form any neighbour tried. Copy maxwhite to memO
~nd repeat to get maxwhite for 16 pixels around current pixel.
Copy result to memO when convolution complete.
3. Do large neighbourhood convolution for pixels where the small neighbourhood convolution gives a poor shading correction.
'poorwht' defines the number of white states tout of 32! which are in the 'poor' category. 'poorwht' is set in hstgr; for clearly black/white im~ges it is 2, for images which have some range of grays it is 7, and for images which have the histogram of grayscale image it is 12. Erase overlay 8 when doing the final imsge move to set up for step 4.

4. Sum maxwhite for 4 pixels- 3 nearest neighbours and current pixel - -except at edge of image. Feed result to lookup table to get shad-ing correction.

5. Now correct imsge by adding correction to video.
Return result to mem),6. invert in O if needed.
-3~

lZ~33477 based bO
half bufptr bufptr2 bufptr4 bufptr6 bufptr8 bufptrlO
bufputl2 bufptrl4 bufptrl6 bufptrl8 bufptr20 bufptr22 bufptr24 bufptr26 bufptr28 bufptr30:
h~lf xd(b~se b2) yd(baseb2+12) md(base b2+24);
statlc half xs(6) ys(6) ms(6) x1(6) yl(6) ml(6) sf(6) yf(6) mf(6);
init xs _2 _1 2 _4 1 5; (x,y,ms for maxwhite selection) init ys 2 _2 1 _2 _S 3;
init ms O O 1 0 0 2; (copy 6 to O if ms=l, exit if ms=2) init xl _28 _12 60 28 0 0 init yl 44 _52 _4 _32 0 0;
init ml O O O 1 0 0;
init xf 1 _1 0 0 0 0; (x,y,mf for final averaging) init yf _l _l 1 0 0 0;
init mf O O 1 1 1 1; (exit when mf =0) bO=addr omdbuf bOst=bO
if color=O
tvarO= eor 255 max (note that inversion interchanges max and min) med= eor 255 med (invert median, meaximum and minimum) max= eor 255 min min= tvarO
bufptr = hex aO10 bufptr2 = hex c800 bufptr4 = hex caOO (memO to 0,1,2,3) bufptr6 = hex cc36 (clamp 2,3, invert 0,1) bufptr8 = hex 8081 (enable 0,7) bufptrlO= hex aOOO (write the inverted image to 0,7) bufptrl2= hex aOOl (select vidoe driver) bufptrl4= hex ca28 (invert image display to outputs a,c) bO= +bO 16 entif zerotbl 5 (tblO,2: zero hi8h order input tables) shdtbl5 (tbll: 5 bit wite covering range med+/-dif) bufptr = hex aOOl (select video driver per bitO only) bufptr2 = hex c837 (video a from 7, b from 3, c from 7) bufptr4 = hex aO10 bufptr6 = hex aO10 bufptr8 = hex ca30 (tbl4 on, carry to overlay) bufptrlO = hex cc34 (memO only, thru tbl 1) bufptrl2 = hex 8341 (enable memO,6,8,9) bufptrl4 = hex cOOO (5 bit white to memO,6, clear overlays 8,9) bO=+bO 16 (set up for convolution with nearest neighbours) mtable 1 _3 (tblO; shift high order input down 3 bits) setbits (set bits to flag poor white valueY) mtable 2 5 (tbll: shift low order input up 5 bits) zerotbl 4 (tbl2: zeroed) ~ 4~

lZ83477 setbits (set bits to flag poor white values) mxwhite4 (tbl4: table to select maximum white) (tbl3 is still 1:1) b2zb9 rtn=writgl bOst wxpl -bO bOst O rcadd (write instructions to Grinnell) lf rtn-=O stop = sll color 1 1 endif (Select max white for 5 bit states in memO, mem6 via convolution) bO=bOst b2=addr xs begin (high-pass maxwhite selection) bufptr = hex aOO9 (select zoom and pan) bufptr2 = hex bOOO (point to x offset) bufptr4 = bex cOOc (pan enabled, zoom=l, blanking at edge) bufptr6 =~hex dOff xd (set x offset) bufptr8 =+hex dOff (set y offset) bufptrlO= hex aO10 (select processor) bufptrl2= hex c8fO (memO to tbls 0,1; mem6 to tbls 2,3) bufptrl4= hex ca30 (tbl4 on, carry to overlay) bufptrl6= hex cc30 (add maxwhite to shifted maxwhite) bufptrl8= hex 8140 (enable mem6, overlay 8) bufptr20= hex cOOO (write to mem6, overlay 8) bO= + bO 22 if md- o (repeat operation with new neighbour) b2--+ b2 2 repeat else (copy maxwhite in mem6 to memO) bufptr = hex aO08 (select zoom and pan) bufptr2 = hex cOOO (disable zoon amd pan) bufptr4 = hex aO40 (enable digitizer 0) bufptr6 = hex 8001 (select memO) bufptr8 = hex c80f (input from 6, camera input zeroed) bufptrlO= hex d200 (copy 6 to 0) bO= + bO 12 if md=l b2=~ b2 2 repeat (repeat unitl md=2) endif endbegin (leave loop hwen md=2) b2=addr xl begin (low-pass maxwhite selection) bufptr = hex aO08 (select ~oon and pan) bufptr2 = hex bOOO (point to x offset) bufptr4 = hex cOOc (pan enabled, zoom=l, blanking at edge) bufptr6 =+hex dOff xd (set x offset) bufptr8 =+hex dOff yd (set y offset) bufptrlO= hex aO10 (select processor) bufptrl2= hex c8fO (memO to tbls 0,1; mem6 to tbls 2,3) bufptrl4= hex call (tbll4 on, processor toggling on) bufptrl6= hex cc31 (keep maxwhite as-is if overlay 8 not set) bufptrl8= hex ce30 (add maxwhite to shifted maxwhite if ovly set) bufptr20~ hex 8040 (enable mem6) lZ83477 bufptr22= hex cOOO (write to mem6) bO= + bO 24 if md= O (repeat operation with new neighbour) b2=+ b2 2 repeat endif endbegin (leave loop when md-=O) b2=b9 rtn=writgl bOst expl -bO bOst O rcadd (write instructions to Grinnell) if rtn~=O stop - sll colour 1 1 endif bO--bOst zerobtl 5 (zero high order input tables) (move image from mem6 to memO, clearing overlay 8 as well) bufptr = hex ~008 (select zoom and pan) bufptr2 = hex cOOO (disable zoom and pan) bufptr4 = hex aOlO (select processor) bufptr6 = hex c8ff (mem6 to tbis 0,1,2,3) bufptr8 = hex ca30 ttbl4 on, carry to output) bufptrlO= hex cc31 (add, clamp 0,1) bufptrl2= hex 8101 (enable memO, overlay 8) bufptrl4= hex cOOO (copy 6 to 0, clear overlay 8) bO=+bOl6 tbll21 hex la (make tbls 1,3 1st 1/4 of tbl4 1:1) shdtblx4 (3rd quadrant of tbl4 4x for border pixels) (set overlay masks to block averaging of edge pixels) bufptr = hex 8300 (select overlays 8,9) bufptr2 = hex 1300 (select subchannels 8,9) bufptr4 = hex 2800 (set normal write mode) bufptr6 = hex 4800 (Ea-O) bufptr8 = hex 51ff (Eb=lll) bufptrlO= hex 6800 (La=O) bufptrl2= hex 7400 (Lb=O, write bottom line of masks) bufptrl4- hex 4800 (Ea=O) bufptrl6= hex 51ff (Eb=511) bufptrl8- hex 69df (Las479) bufptr20= hex 7400 (Lb=O, write top line of masks) bufptr22= hex 4800 (Ea=O) bufptr24= hex SOOO (Eb=O) bufptr26= hex 6800 (La=O) bufptr28= hex 75df (Lb=479, write right side of masks) bO-+bO 30 bufptr = hex 49ff tEa=5ll) bufptr2 = hex 5000 (Eb=O) bufptr4 = hex 6800 (La=O) bufptr6 = hex 75df (Lb=479, wirte right side of masks) bO=~bO 8 b2=addr xf begin (average 4 maxwhite values, keep edge as-is) bufptr = hex aO08 (select zoom and pan) bufptr2 = hex bOOO (point to x offset) bufptr4 = hex cOOc (pan enabled, zoom=l, blanking at edge) bufptr6 =+hex doff xd (set x offset) ~L2 . . .

lZ83477 bufptr8 =+hex dOff yd (set y offset) bufptrlO= hex aO10 (select processor) bufptrl2= hex c8fO (memO to tbls 0,1; mem6 to tbls 2,3) bufptrl4= hex cc30 (add if overlay not set) bufptrl6= hex ce31 (2,3 only if overlay set no averaging) lf md=O
bufptrl8= hex call (tbl4 on, proc tog on, tbl tog off) bufptr20= hex 8040 (enable mem6) bufptr22= hex cOOO (write to mem6, overlay 8) bO=+bO 24 b2=+b2 2 repeat else (sum of 4 maxwhite, normalize edges to 4x) bufptrl8= hex ca35 (tbl4 on, proc & tbl tog on, carry to ovly) bufptr20= hex 8340 (enable mem6, ovly 8.8 &9.9) bufptr22= hex cOOO (write to mem6, clear overlays) bO=+bO 24 endif endbegin (now rebuild max white via tbls 0,1. Original image in 7, shading correction in 6. Tbl3 still 1:1, tbl 2 zeroed) shdtblx (maxwhite toshsding correction in 0,1) clip4 (tbl4 1:1 for real video lnput range) bufptr = hex aO08 (sslect zoom and pan) bufptr2 = hex cOOO (disable zoom and pan) bufptr4 = hex aO10 (select processor) bufptr6 = hex c8ff (mem6 to 0,1) bufptr8 = hex cadO (mem7 to 2,3, tbl4 selected) bufptrlO= hex cc30 (add 0,1 to 2,3) bufptrl2= hex 80cl (memO,6,7 enabled) bufptrl4= hex cOOO (write to memO,6; clear mem7) bufptrl6= hex aOOl (select video driver) . -bufptrl8= hex c804 (restore output a,c to memO) bO= +bO 20 tbll21 hex Of (all input tables) lf colour=O
tvarO = eor 255 max (note inversion interchanges max, min) med= eor 255 med max= oer 255 min min= tvarO
bufptr = hex aO10 (select processor) bufptr2 = hex c800 (memO to 0,1,2,3) bufptr4 = hex caOO (deselect tbl4) bufptr6 = hex cc36 ~clamp 2,3, invert 0,1) bufptr8 = hex 8041 (enable memO,6) bufptrlO= hex cOOO (write the inverted image to 0,6)bufptrl2= hex aOOl (select video driverO) bufptrl4= hex caOO (resotre normal output from inverted) bO--+bO 16 endif b2=b9 rtn=writgl bOst expl -bO bOst O rcadd(srite instructions to Grlnnell) .. ~ ~ . .

Y09-82-040X ~X83477 if rtn~=O stopO endif endfn (end of shading correction function) function setbits:
('or' flrst poorwht table vslues with hex 80 to flag low intensities) bsubO=bO (save instruction pointer) bO=-bO 258 (back up to start of table) bl=+bO expl poorwht (set limit for loop) begin (set msbs for poorwht entries) if bO llt bl bytptr= or bytptr hex 80 bO ~+ bO 1 repeat endif endbegin bO=bsubO (restore instruction pointer) endfn function shdtblS;
(Generate table for medlan-video, limited to a maximum of dif and scaled to fit in S bits. Offset table output such taht min white i8 ), max white is 31. Generate in packed form.) bufptr = hex aO10 (select processor) bufptr2=hex c222 (write to table 1) bufptr4=hex bOOO (zero address counter) bufptr6-hex a210 (select byte unpacking too) bufptr8-hex c480 (download 256 packed bytes) bO=+ bO 10 ln=-med dif (min limit) lp=+med dif (max limit) if ln~O (zero output to video input of ln) zero bO expl ln bOS+ bO expl srl dif 1 endif bl-expl srl dif 1 b2~0 b3sO
be8in if b2 bl blS+ bl expl dif (incr in steps of dif) if b3 31 (limit b3 to S bits or less) b3s+ b3 1 endif endif bytptr=expl b3 b2=~ b2 16 (incr in steps of 16. equiv to incr of ln by 1) bOS+ bO 1 if ln<lp ln=~ ln 1 repeat endif endbegin 4~

12834~7 lf lp<255 flll bO 31 expl -255 lp (flll at bO with 31 for 255-lp bytes) bO=+ bO expl -255 lp else bO =- bO expl -lp 255 (back up if any overshoot) endif bufptr = hex aO10 (deselect byte unpacking) bO=+ bO 2 endfn function mxwhite4;
(Wrlte 10 x 8 max white select table to processor table4.
The two white values are in hte low and high 5 bit sections of the input, unsigned, with values O to 31, each.
Wrlte out in packed form.) static byte databl(32);
lnit databl 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31;
bufptr =hex aO10 (select processor) bufptr2-hex c230 (secect table 4) bufptr4=hex bOOO (zero-address counter) bufptr6--hex a210 (select bytepacking) bufptr8--hex c400 (download 512 bytes in packed form) bO-+ bO 10 bsubO--bO
bl=bO
b2=addr datatbl copy bO b2 32 (copy initial buffer) bO=+bO 32 copy bO b2 32 (twice) bO=+bO 32 b2GbsubO
copy bO b2 b4 (replicate 1st 64) bO=+bO 64 copy bO b2 128 (replicate 1st 128) bO=~bO 128 copy bO b2 256 (complete 1st 512 table entries) bO=~bsubO 32 tback up bO to 1st fill point) (bO points to fill srea in loop) bl=l (counter for high order white) begin (do 1st half of table4) fill bO expl bl expl bl (fill in bl for bl entrles) bO=~bO 32 bl=~ bl 1 if bl~l6 repeat endif endbegin bufptr--hex c400 (download 256 words in packed form) bO=~ bO 2 ~5 1~8~477 copy bO bsub 512 (copy 1st half of table to 2nd half) begin (do 2nd half of table4) fill bO expl bl expl bl (overwrite where needed) bO=+ bO 32 if bl<31 bl=+ bl 1 repeat endif endbegin bufptr = hex aO10 (deselect byte unpacking) bO=+ bO 2 endfn function shdtblx4;
(make 3rd quadrsnt of tbl4 4x) bufptr = hex aO10 (select processor) bufptr2 - hex c230 (select table 4 only) bufptr4 = hex b200 (set address at upper half of table) bufptr6 = hex a210 (select bytepacking) bufptr8 = hex c480 (downlosd 256 bytes in apcked form) bO=+bO 10 bsubO=+bO 256 bl-hex 0004 (~nit start of tablt) b2-hex 0808 begin (4x tsble) bufptr=expl bl bO=+bO 2 if bO llt bsubO
bl=~bl expl b2 repest endif endbegin bufptr=hex aO10 (deselect byte unpacking) bO=+bO 2 endfn function shdtblx;
(Gener8te tables for medi8n-video, expanding sum of four 5 bit m8xwhite values to true difference (dif). High order table should be Of for negative output. Output is added to video to correct for shading. Generate in packed form.
Write to tables 0,1.) bufptr = hex aO10 (select processor) bufptr2 = hex c222 (select table 1) bufptr4 = hex bOOO (zero address) bufptr6 = hex a210 (select bytepacking) bufptr8 = hex c480 (download 256 bytes in packed form) bO=+ bO 10 bsubO~bO 128 bl=expl - sll dif 6 dif (init output of table at 63 x diff) be8in (generate rest of 32 output states) bO =~ bO 1 if bO llt bsubO

~G

lX834'77 bl=- bl expl dif repeat endif endbegin (last output is _dif) fill bO expl _dif 128 (fill out rest of table with _dif) bO=+ bO 128 bufptr = hex aO10 (select processor again) bufptr2 = hex c221 (write totable 0) bufptr4 = hex bOOO (zero address) bufptr6 = hex a210 (select bytepacking) bufptr8 = hex c480 (download 256 bytes in packed form) bO=+ bO 10 zero bO 64 (first 64 output states are positive) bO=+ bO 64 fill bO 15 192 (set sign bits for rest) bO=+ bO 2 endfn function mtable msk(h) s(h);
(write processor tsbles specified by msk - download in packed form) (if to table 4, downlaod 512 bytes, rest take 256) bufptr -hex aO10 if msk<O (if msk negative, create signed table) msk=_msk bufptr2=+hex c220 msk (write to tables set in msk) bufptr4=hex bO80 (address conter-_128) bl-_128 else bl=O
bufptr2=+hex c220 msk (write to tables set in msk) bufptr4=hex bOOO (address counter=O) entif bufptr6=hex a210 (select processor and byte unpacking) bufptr8=hex c480 (download 128 packed words) bsubO=+bO 266 (256 + init sequence length) if msk=hex 10 (special case for table 4) if ~<0 (expecting more than 8 bits, 90 extend divide table) bl= expa sll expa bl l (double tbl length, so double bl) if bl < O (signed divide) bufptr4=hex b300 (start at _ 256 address) endif bufptr8=hex c400 (download 256 packed words) bsubO=+bO 522 (512 + init sequence length) endif endif bO=+ bO 10 (init sequence length) if s~O (multiply) begin bytptr= expl and 255 sll expl bl s bO=+ bO 1 if bO llt bsubO

~7 1;~8347~

bl=+ bl 1 repeat endif endbegln else (divide) 8-_8 begin bytptr= expl and 255 srl expl bl s bO=+ bO 1 lf bO llt bsubO
bl=+bl 1 repeat endif endbegin endif bufptr= hex zO10 (deselect byte unpacking) bO=+ bO 2 endfn function zeorbtl msk(h);
(zero the table defined by msk. Output in apcked form) bufptr =hex aO10 bufptr2=+hex c220 msk (write to tables set in msk) bufptr4-hex bOOO (zero address conter) bufptr6-hex a210 (select byte unpacklng too) bufptr8=hex c480 (download 256 bytes in packed form) bO--+ bO 10 zero bO 256 (fill the table with zeros) bO=+ bO 256 bufptr=hex aO10 (deselect byte unpacking mode) bO=+ bO 2 endfn function tbll21 msk(h);
(Write 1:1 processor tables specified by msk - download in packed form) ~If to table 4, download to quadrants specified by hi8h order byte of msk. the first nibble gives the startin8 address, the second glves the final address. If both are zero, do the zeroth quadrant only.
the table loading must be contiguous. Thus, to download all four quadrants of tbl4. msk should be hex 4010.) bufptr = hex aO10 (select processor) bufptr2=+hex c220 and ehx OOlf msk (write to tables set in msk) bsubl=expa srl and msk hex fOOO 4 (end of table - may be zero) msk=and hex OfOO msk (address of start of table) bufptr4=hex a210 (select both processor and byte unpacking) bufptr6=or ehx bOOO msk (starting table address) bufptr8 = hex c480 (downlaod 128 packed words) bO=~ bO 10 (init sequence length) bsubO=+bO 256 (point to end of 1st sectlon) bl=l b2=hex 0202 be8in bufptr= expl bl bO=+ bO 2 ~283477 if bO llt bsubO
bl=+bl expl b2 repeat endif endbegin msk=+msk hex 100 (point to address of current end of table) if bsubl>expl msk (real end of table > current end of table) bl=-bsubl expl msk (address differnece) bl=+bl srl expl bl 7 (total number of bytes to copy) movechar bO -bsubO 258 expl bl (replicatel:l table) bO=+bO expl bl endfif bufptr = ehx aO10 (deselect bytepacking) bO=+bO 2 endfn function clip4;
(clip to real video in table4. Generate table4 in packed form) tbll21 hex 10 (lst quadrant of table 4) bufptr = hex a210 (select bytepacking) bufptr2= hex c480 (start second quadrant) bO=+ bO 256 bufptr-hex c400 (download 512 bytes in packed form) bO=+ bO
fill bO hex 00 512 (clamp video below O to O) bO=+ bO 512 bufptr=hex aO10 (deselect byte unpacking) bO=+ bO 2 endfn ~7

Claims (13)

1. A method for correcting shading effects in a video image using a histogram of gray scale intensity values of picture elements or pels making up the image, which histogram contains a median and black and white color extremes, comprising the steps of:
generating a histogram of distribution of gray scale intensity values of the pels in an original image to be corrected:
determining an acceptable range of background intensity values by:
calculating the median and the black and white color extremes in the histogram: and calculating the distance between the histogram median and the color extreme on the side of the median having the majority of gray scale intensity values and setting said range as twice that distance centered at the median;
creating a first background image by sampling the gray scale intensity value of each pel to be corrected and setting all values outside of said acceptable range to that of the color extreme least distant therefrom;
creating a duplicate background image from said first background image;
sampling the gray scale values of a number of pels in a predetermined region around a first pel in said duplicate background image;

changing the value of the pel in said first background image corresponding to said first pel, to the largest value among those of said first pel and the sampled pels;
continuing the foregoing sampling and changing steps until all of the pel values in said first background image have been changed accordingly;
changing the duplicate background image in accordance with the changes in said first background image;
repeating the sampling step over a larger region and a larger number of pels with the corresponding changing in the first background image; and repeating the foregoing changing until a desired corrected background image is obtained which becomes the shading correction for the original image.

2. A method as in claim 1 further comprising the steps of carrying out a further correction while sampling by detecting changed pel values below a given value and storing indications thereof; and using said indications to cause only said below value pels to be corrected during subsequent sampling and changing.

3. A method as in claim 2 wherein said method is carried out in a computer and said indications are stored in an overlay plane, which indications are changed during successive samplings accordingly to provide a check as to the adequacy of the corrected image.

4. A method for correcting shading effects in a video image using a histogram of gray scale intensity values of picture elements or pels making up the image, which histogram contains a median and black and white color extremes, comprising the steps of:
generating a histogram of distribution of gray scale intensity values of the pels in an original image to be corrected;
determining an acceptable range of background intensity values by:
calculating the median and the black and white color extremes in the histogram; and calculating the distance between the histogram median and the color extreme on the side of the median having the majority of gray scale intensity values and setting said range as twice that distance centered at the median; and creating a correction image by:
comparing the gray scale intensity value of each pel to be corrected in the original image with the median value and calculating respective difference values;
correcting the value of each said pel by its respective difference value when the latter is within said range; and not correcting the value of each said pel when its respective difference value is outside of said range.

5. A method as in claim 4 wherein said original image is remapped to said correction image using a lookup table, the output of which is said respective difference values.

6. A method as in claim 4 comprising the further steps of:
sampling the gray scale values of a number of pels in a predetermined region around a first pel in said correction image;
summing said sampled values and dividing the sum by the number of samples to obtain an average value and using said average value to correct the value of said first pel; and continuing the foregoing summing and correcting steps until all of the pels in said correction image have been corrected accordingly;

7. A system for correcting shading effects in a video image using a histogram of gray scale intensity values of picture elements or pels making up the image, which histogram contains a median and black and white color extremes, comprising:
means for generating a histogram of distribution of gray scale intensity values of the pels in an original image to be corrected;
means for determining an acceptable image of background intensity values comprising: means for calculating the median and the black and white color extremes in the histogram; and means for calculating the distance between the histogram median and the color extreme on the side of the median having the majority of gray scale intensity values and setting said range as twice that distance centered at the median;

means for creating a first background image by sampling the gray scale intensity value of each pel to be corrected and setting all values outside of said acceptable range to that of the color extreme least distant therefrom;
means for creating a duplicate background image from said first background image;
first means for sampling the gray scale values of a number of pels in a predetermined region around a first pel in said duplicate background image:
first means for changing the value of the pel in said first background image corresponding to said first pel, to the largest value among those of said first pel and the sampled pels;
means for reactivating said first sampling and changing means until all of the pel values in said first background image have been changed accordingly;
second means for changing the duplicate background image in accordance with the changes in said first background image;
second means for sampling the gray scale values in said duplicate image over a larger region and a larger number of pels than said first sampling means and activating said first changing means to make the corresponding changes in the first background image; and means for activating said second changing means until a desired corrected background image is obtained which becomes the shading correction for the original image.

8. A system as in claim 7 further comprising:

means for carrying out a further correction while sampling by detecting changed pel values below a given value and storing indications thereof, and means for using said indications to cause only said below value pels to be corrected during subsequent sampling and changing.

9. A system as in claim 8 wherein said carrying out means comprises:
computer means for storing said indications in an overlay plane, and for changing said indications during successive samplings accordingly to provide a check as to the adequacy of the corrected image.

10. A system for correcting shading effects in a video image using a histogram of gray scale intensity values of picture elements or pels making up the image, which histogram contains a median and black and white color extremes, comprising:
means for generating a histogram of distribution of gray scale intensity values of the pels in an original image to be corrected;
means for determining an acceptable range of background intensity values comprising:
means for calculating the median and the black and white color extremes in the histogram; and means for calculating the distance between the histogram median and the color extreme on the side of the median having the majority of gray scale intensity values and setting said range as twice that distance centered at the median; and means for creating a correction image comprising:
means for comparing the gray scale intensity value of each pel to be corrected in the original image with the median value and calculating respective difference values;
means for correcting the value of each said pel by its respective difference value when the latter is within said range; and means for not correcting the value of each said pel when its respective difference value is outside of said range.

11. A system as in claim 10 further comprising means for remapping said original image to said correction image using a lookup table, the output of which is said respective difference values.

12. A system as in claim 10 further comprising:
means for sampling the gray scale values of a number of pels in a predetermined region around a first pel in said correction image;
means for summing said sampled values and dividing the sum by the number of samples to obtain an average value;
means using said average value correcting the value of said first pel; and means for reactivating said summing means and said correcting means until all of the pels in said correction image have been corrected accordingly.

13. A method for correcting shading effects in a video image using a histogram of gray scale intensity values of picture elements or pels making up the image, which histogram contains a median and black and white color extremes, comprising the steps of:
generating a histogram of distribution of gray scale intensity values of the pels in an original image to be corrected;
determining an acceptable range of background intensity values by:
calculating the median and the black and white color extremes in the histogram; and calculating the distance between the histogram median and the color extreme on the side of the median having the majority of gray scale intensity values and setting said range as twice that distance centered at the median; and creating a correction image by:
comparing the gray scale intensity value of each pel to be corrected in the original image with the median value and calculating respective difference values;
correcting the value of each said pel by its respective difference value when the latter is within said range.