Traitement en cours

Veuillez attendre...

Paramétrages

Paramétrages

Aller à Demande

1. WO1993011629 - PROCEDE ET SYSTEME DE COMPRESSION D'IMAGE DE DOCUMENT

Note: Texte fondé sur des processus automatiques de reconnaissance optique de caractères. Seule la version PDF a une valeur juridique

[ EN ]

WHAT IS CLAIMED IS:
1. In a discrete transform system for compressing document image data arranged as a plurality of pixels, each representing one of a plurality of gray levels, the transform including a plurality of transform coefficients, the improvement comprising:
means for estimating a packet size of memory space required to store the document image data for a given document after compression by the discrete transform technique; and
selection processor means coupled to the means for estimating and operative to select one of a plurality of matrices of transform coefficient modifiers as a function of a packet size estimate and to transmit the selected one matrix of transform coefficient modifiers for the given document to a transform compressor for use in altering the plurality of transform coefficients.

2. The improvement of claim 1 further comprising means for examining each pixel and generating a histogram of all pixels in the document image data;
histogram processing means for generating a contrast reduction pixel conversion function and a gray level stretch pixel conversion function in accordance with preselected characteristics of the histogram; and
means for combining the contrast reduction and gray level stretch conversion functions and applying a combined conversion function to the document image data and for transmitting converted image data to the means for estimating packet size.

3. The improvement of claim 1, wherein the means for estimating comprises processor means for generating a running sum of absolute values of the plurality of transform coefficients.

4. The improvement of claim 3 , wherein the discrete transform comprises a discrete cosine transform as defined by the Joint Photographic Experts Group Of The International Standards Organization.

5. The improvement of claim 1, wherein the means for estimating comprises a transform compressor utilizing a preselected test matrix of transform coefficient modifiers.

6. The improvement of claim 2, wherein the contrast reduction pixel conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope less than unity for input pixel gray level values less than a predetermined contrast reduction protection value selected as a function of characteristics of the histogram, and wherein a converted output pixel gray level value is equal to a corresponding input pixel gray level value for input pixel gray level values equal to or greater than the contrast reduction protection value.

7. The improvement of claim 2, wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray level values less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

8. The improvement of claim 6 , wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray level values less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

9. The improvement of claim 8, wherein the contrast reduction pixel conversion function and the gray level stretch conversion functions are combined and implemented as a look-up table.

10. The improvement of claim 6, wherein the contrast reduction protection value is a predetermined percentage of a gray level value corresponding to a preselected peak value of the histogram.

11. The improvement of claim 7, wherein the black clip value is determined by comparing a cumulative sum of histogram frequencies to a first preselected threshold value and a white clip value is determined by comparing a cumulative sum of histogram frequencies to a second preselected threshold value.

12. In a discrete transform compression method for compressing document image data arranged as a plurality of pixels, each representing one of a plurality of gray levels, wherein a plurality of transform frequency coefficients corresponding to the pixels are generated by a preselected transform function and the coefficients then quantized by a matrix of quantization values, the improvement comprising, for each document whose image data representation is to be compressed:
a) estimating a packet size of memory space required to store the document image data after compression by the discrete transform technique; and
b) selecting one of a plurality of matrices of quantization values, as a function of a packet size estimate, to be applied to transform frequency coefficients generated by the preselected transform function.

13. The improved method of claim 12, wherein the plurality of matrices are generated from a training set comprising a plurality of typical documents whose image data are to be compressed.

14. The improved method of claim 12, wherein the selection of the one of the plurality of matrices of quantization values is performed so as to maintain an average packet size of a series of document images within a predefined range of acceptable values.

15. The improved method of claim 12, further comprising preliminary steps of:
(i) generating a histogram of all pixels in the document image data;
(ii) generating a contrast reduction pixel conversion function and a gray level stretch conversion function in accordance with preselected characteristics of the histogram; and (iii) combining the contrast reduction and gray level stretch conversion functions and applying a combined conversion function to the document image data prior to estimating the packet size.

16. The improved method of claim 12, .wherein a packet size is estimated by generating a running sum of absolute values of the transform frequency coefficients.

17. The improved method of claim 12, wherein a packet size is estimated by actually compressing the document image data with the discrete transform compression method utilizing a preselected test matrix of quantization values.

18. The improved method of claim 12, wherein step b) comprises:
1) compressing the document image data
utilizing a predetermined reference
quantization matrix to obtain a reference
document image packet size;
2) determining a target document image packet
size utilizing a preselected transfer
function applied to the reference document
image packet size;
3) calculating, for each of the plurality of
matrices of quantization values, an
estimated document Image packet size as a
function of a number of pixels per document
and the reference document image packet
size; and
4) selecting a matrix of quantization values
which yields an estimated document image
packet size closest to, but not exceeding,
the target document image packet size .

19. The improved method of claim 15, wherein the contrast reduction pixel conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope less than unity for input pixel gray level values less than a predetermined contrast reduction protection value selected as a function of characteristics of the histogram, and wherein a converted output pixel gray level value is equal to a corresponding input pixel gray level value for input pixel gray level values equal to or greater than the contrast reduction protection value.

20. The improved method of claim 15, wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray levels less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

21. The improved method of claim 19, wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray level values less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

22. The improved method of claim 21, wherein the contrast reduction pixel conversion function and the gray level stretch conversion function are combined and implemented as a look-up table.

23. The improved method of claim 19, wherein the contrast reduction protection value is a predetermined percentage of a gray level value corresponding to a preselected peak value of the histogram.

24. The improved method of claim 20, wherein the black clip value is determined by comparing a cumulative sum of histogram frequencies to a first preselected threshold value and the white clip value is determined by comparing a cumulative sum of histogram frequencies to a second preselected threshold value.

25. The improved method of claim 23, wherein the slope of the linear function for pixel gray level values less than the contrast reduction protection value is empirically determined for a test set of documents of a class of documents to be compressed.

26. The improved method of claim 12, wherein the discrete transform comprises a discrete cosine transform as defined by the Joint Photographic Experts Group Of The International Standards Organization.

27. In a discrete transform system for compressing document image data arranged as a plurality of pixels, each representing one of a plurality of gray levels, the transform including a plurality of transform coefficients, the improvement comprising:
means for examining each pixel and generating a histogram of all pixels in the document image data;
histogram processing means for generating a contrast reduction pixel conversion function and a gray level stretch pixel conversion function in accordance with preselected characteristics of the histogram;
means for applying the contrast reduction and gray level stretch conversion functions to the document image data;
means for estimating a packet size of memory space required to store the document image data for a given document after compression by the discrete transform technique, the means for estimating coupled for receipt of the converted image data from the histogram processing means; and
selection processor means coupled to the means for estimating and operative to select one of a plurality of matrices of transform coefficient modifiers as a function of a packet size estimate and to transmit the selected one matrix of transform coefficient modifiers for the given document to a transform compressor for use in altering the plurality of transform coefficients.

28. In a discrete transform compression method for compressing document image data arranged as a plurality of pixels each representing one of a plurality of gray levels, wherein a plurality of transform frequency coefficients corresponding to the pixels are generated by a preselected transform function and the coefficients then quantized by a matrix of quantization values, the improvement comprising, for each document whose image data representation is to be compressed.

a) generating a histogram of all pixels in the document image data;
b) generating a contrast reduction pixel conversion function and a gray level stretch .conversion function in accordance with preselected characteristics of the histogram;
c) applying the contrast reduction and gray level stretch conversion functions to the document image data;
d) estimating a packet size of memory space required to store the document image data after compression by the discrete transform technique; and
e) selecting one of a plurality of matrices of quantization values as a function of a packet size estimate to be applied to transform frequency coefficients generated by the preselected transform function.

29. Pre-processing apparatus for document image data arranged as a plurality of pixels, each representing one of a plurality of gray levels, the apparatus comprising:
histogram processing means for generating a contrast reduction pixel conversion function and a gray level stretch pixel conversion function in accordance with preselected characteristics of the histogram; and
means for combining the contrast reduction and gray level stretch conversion functions and applying a combined conversion function to the document image data.

30. The apparatus of claim 29, wherein the contrast reduction pixel conversion function comprises a piecewise linear function wherein a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope less than unity for input pixel gray level values less than a predetermined contrast reduction protection value selected as a function of characteristics of the histogram, and wherein a converted output pixel gray level value is equal to a corresponding input pixel gray level value for input pixel gray level values equal to or greater than the contrast reduction protection value.

31. The apparatus of claim 29, wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray level values less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than "unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

32. The apparatus of claim 30, wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray level values less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

33. The apparatus of claim 32, wherein the contrast reduction pixel conversion function and the gray level stretch conversion functions are combined and implemented as a look-up table.

34. The apparatus of claim 30, wherein the contrast reduction protection value is a predetermined percentage of a gray level value corresponding to a preselected peak value of the histogram.

35. The apparatus of claim 31, wherein the black clip value is determined by comparing a cumulative sum of histogram frequencies to a first preselected threshold value and a white clip value is determined by comparing a cumulative sum of histogram frequencies to a second preselected threshold value.

36. A method for pre-processing document image data arranged as a plurality of pixels, each representing one of a plurality of gray levels, the method comprising the steps of:

(i) generating a histogram of all pixels in the document image data;

(ii) generating a contrast reduction pixel conversion function and a gray level stretch conversion function in accordance with preselected characteristics of the histogram; and
(iii) combining the contrast reduction and gray level stretch conversion functions and applying a combined conversion function to the document image data.

37. The method of claim 36, wherein the contrast reduction pixel conversion function comprises a piecewise linear function wherein a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope less than unity, for input pixel gray level values less than a predetermined contrast reduction protection value selected as a function of characteristics of the histogram, and wherein a converted output pixel gray level value is equal to a corresponding input pixel gray level value for input pixel gray level values equal to or greater than the contrast reduction protection value.

38. The method of claim 36, wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray levels less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

39. The method of claim 37, wherein the gray level stretch conversion function comprises a piece-wise linear function wherein a converted output pixel gray level value is a blackest value for all input pixel gray level values less than or equal to a predetermined black clip value selected as a function of characteristics of the histogram, a converted output pixel gray level value is related to a corresponding input pixel gray level value by a linear function having a slope greater than unity for input pixel gray level values greater than the black clip value and less than a predetermined white clip value selected as a function of characteristics of the histogram, and a converted output pixel gray level value is a whitest value for all input pixel gray level values greater than or equal to the white clip value.

40. The method of claim 39, wherein the contrast reduction pixel conversion function and the gray level stretch conversion function are combined and implemented as a look-up table.

41. The improved method of claim 37, wherein the contrast reduction protection value is a predetermined percentage of a gray level value corresponding to a preselected peak value of the histogram.

42. The method of claim 38, wherein the black clip value is determined by comparing a cumulative sum of histogram frequencies to a first preselected threshold value and the white clip value is determined by comparing a cumulative sum of histogram frequencies to a second preselected threshold value.

43. The method of claim 41, wherein the slope of the linear function for pixel gray level values less than the contrast reduction protection value is empirically determined for a test set of documents of a class of documents to be compressed.

APPENDIX 1

/* Main */
read_hist_fϊle (hist_file, hist, ingrays); /* Read histogram data */
get_CRP (hist, ingrays, CRP); /* Compute CR protected range */ get_control_points (hist, ingrays, bcp, wcp); /* Get stretch control points */ make_comb_LUT (comblut, ingrays, outgrays, bcp, wcp, CRP);

APPENDIX 2

/* get _ control_points */
#defme BLK_INT_THRESH 1000
#defϊne WHT_l NT_ THRESH 1000
get_control_points (hist, ingrays, bcp, wcp);
sum = 0;
i = 0;
while (hist[i] = =0) /* Skip to the first non-zero */ i+ + ; /* histogram entry */ i+ + ; /* Skip this entry because it */
/* contains the overscan */ while (sum < BLK_INT_THRESH) { /* Integrate histogram until */ sum = sum + histfi]; /* frequency threshold is reached.*/ i+ + ;
}
bcp = i; /* This is black control point */ sum = 0;
i = ingrays-1;
while (sum < WHT_INT_THRESH) { /* Integrate histogram unit */ sum = sum + hist[i]; /* frequency threshold is reached. */ i--;
}
wcp = i; /* This is white control point */ return;

APPENDIX 3

/* make_comb_LUT */
make_comb_LUT (comblut, ingrays, outgrays, bcp, wcp, CRP);
bcpcr = cr(bcp, CRP);
for (i=0;i< = bcp; i++)
comblutfi] = 0;
for (i=bcp+1; i< =wcp;i++) /* Equation 4 */
comblut[i] = (int)((outgrays-1)*(double)(cr(i,CRP)-bcpcr)/(wcp-bcpcr) +0.5); for (i = wcp + 1; i<ingrays; i+ +)
comblutfi] = outgrays-1;
return;
/* cr */
#define COMPRESS 2.0
cr(i, CRP);
if (i < CRP) /* Equation 1a */
return((int)((1/COMPRESS)*i + (COMPRESS-1)*CRP/COMPRESS + 0.5)); else
return(i); /* Equation 1b */