Edges in Visual Scenes and Sequences: Application to Filtering, Sampling and Adaptive DPCM Coding

Abstract

Although quite natural in scene analysis, the consideration of local edge occurrence and of picture structure implying a certain kind of nonstationarity has not been used much within the framework of picture coding. Some years ago, mainly in order to analyse coding impairments on pictures and their perception by the human visual system, we introduced the notion of «local critical configurations)) and this led us to consider pictures as a more or less random concatenation of these configurations each with their own statistical properties [1]. Extension to moving sequences is straightforward, the concatenation being now dynamical and each configuration bearing its own motion parameters. This approach departs clearly from the usual stationary approach the merit of which was the design of the two main coding families : DPCM coding using a local prediction and the quantization of the error signal ; transform coding using quantization of the transform coefficients on a block-by-block basis. But, more efficient coding schemes for bit rate reduction needed to be defined by taking advantage as much as possible of the specific properties of pictures or sequences and of the human visual system. We refer here to [2] for a recent general review on picture coding and to [3] more specifically for a review on sequence coding.