GB1527276A - Image coder-decoder using a matrix transform with weighted contribution of several points of the image to the formation of one point of the transform - Google Patents

Abstract

1527276 Data transmission; facsimile systems SOC ANON DE TELECOMMUNICATIONS 7 Oct 1976 [24 Oct 1975] 41788/76 Heading H4P In known forms of coding, binary representations of images coded by Hadamond or Haar matrix transforms can produce on reconstruction false or stepped contours by juxtaposition of squares formed by N Î N image panels which in extreme instances can mask information content. The present invention reduces such faults and provides better compression of bits as there is a greater probability of certain matrix coefficients being small or zero (i.e. unwanted coefficients are in effect filtered) so that attenuation of false squares and better statistical approximation of the image is achieved. In the invention there is provided a matrix transform system for coding images with weighted contribution of several points of the image to each point of the coded image and for decoding coded images weighted contribution of several points of the coded image to each point of the decoded image is made. The system comprises: (a) a device for sampling binary representations of lines of an image to be coded and forming with said image samples a square matrix associated with the image; (b) means for splitting said square matrix associated with the image into a plurality of first input component matrices of order 3N Î 3N having a central part of order N Î N; (c) multiplying each of said first input component matrices by a first (e.g. horizontal) rectangular coefficient matrix of order N Î 3N having a square central part in which the coefficients are equal to unity and two square lateral parts in which the coefficients are selectively equal to zero and a predetermined factor smaller than unity, and forming first intermediate matrices of order N x 3N; (d) multiplying each of the first intermediate matrices by a second (e.g. vertical) rectangular coefficient matrix of order 3NÎN which is the transpose of the first rectangular coefficient matrix, and forming first output matrices of order NÎN, each of said first output matrices being the transform of the central part of a first input component matrix; (e) means for forming with the first output matrices a square matrix associated with the coded image; (f) splitting the square matrix associated with the coded image into a plurality of second input component matrices of order 3NÎ3N having a central part of order N Î N; (g) multiplying each of the second input component matrices by a third rectangular coefficient matrix of order NÎ3N having a square central part in which the coefficients are equal to unity and two square lateral parts in which the coefficients are selectively equal to zero and a predetermined factor smaller than unity and forming second intermediate matrices of order NÎ3N; (h) multiplying each of the second intermediate matrices by a fourth rectangular coefficient matrix of order 3NÎN which is the transpose of the third rectangular coefficient matrix and forming second output matrices of order N x N, each of the second output matrices being the transform of the central part of a second input component matrix; and (i) means for forming with the second output matrices a square matrix associated with the decoded image. In the embodiment described N = 2 and the 1st, 2nd, 3rd and 4th rectangular coefficient matrices is a Hadamond matrix but Haar matrices are envisaged. Circuit arrangements for performing the various operations are disclosed.