Adaptive quantification based on actual and predicted image data amounts

Abstract

A procedure for quantification of coefficients in a variable data rate image compression system, in the form of blocks and macroblocks, the system including circuitry for transformation of blocks of pixels into blocks of coefficients, circuitry for quantification of these coefficients, circuitry for storage of coded information before transmission, and circuitry for regulation of quantification as a function of the state of the circuitry for storage. The procedure is such that each image is divided into groups of blocks, at least when a so-called current image must be coded in inter mode, a quantifier is determined at the start of the current image, the quantifier being corrected by a correction coefficient recalculated before the quantification of at least one of the groups, the coefficient being a function of the difference between, first, the number of bits produced for all the groups of blocks previously coded and, secondly, a prediction of this number of bits, the prediction being based on the distribution of the bits during the coding of at least one image preceding the current image. The invention is applicable notably in the field of image compression at low data transmission rates.