Adaptive Split for Image Coding

The purpose of this paper is to present an adaptive split procedure for decomposing an image into a set of squares of different size and its application to image coding. The sub-image contained in each square is approximated in the least mean square sense using polynomial functions. Different algorithms have been established in order to use the separability condition when the approximation is evaluated on a rectangular grid. Starting with a certain set of identical size squares covering the image, the approximation is calculated for each of them. Then, different measurements are performed in order to determine the quality of the approximation inside each square. Whenever this quality is not sufficient, the analysed square is split into four sub-squares of identical size. The same procedure is integrated until the quality is considered good enough. In order to enhance the quality, special smoothing is finally applied on the boundary of big size squares. Examples present the coding efficiency of this type of processing.