Method of and system for coding a two-dimensional data array

Abstract

A two-dimensional data array, such as an orthogonally scanned television image, is divided into a multiplicity of elemental blocks each containing a given number of PCM samples which are converted by linear transformation into a series of coefficients to be quantized for transmission to a remote receiver. The data blocks are processed one at a time by an iterative coder which allots a progressively increasing number of bits to a given block in successive operating stages, the bits being allocated to the several coefficients of the block according to the position of the barycenter of their signal energy assigning the block to one of K classes. An initial overall reconstruction error, computed from the coefficients of a block with no bits yet available for quantization, is updated with each bit allotment in the course of an iteration in which a selected coefficient is subjected to simultaneous coding with S quantizers differing from one another in their quantum steps. The coding operations of the S concurrently active quantizers have different error-reducing effects and the resulting minimum error is compared with a threshold which itself may be variable according to the activity of the image block being processed. When the residual overall error falls below that threshold, processing is terminated and the bits used in the coding of each coefficient are sequentially sent out together with a count of the total number of bits allotted to the block and a parameter k indicating its classification, this information enabling the receiver to reconstruct the image within the limits of the tolerated residual error.