Abstract
An approach to gray-level image coding based on a fractal theory of iterated contractive transformations was first presented by A. E. Jacquin in 1989. To capture the redundancy through self-transformability piecewise, this approach employs spatial image partitioning into range blocks of two sizes with a thresholding operation based on waveform distortion. A set of elemental contractive block transformations from domain blocks to range blocks in an image is referred to as “Iterated Function System (IFS)” since its iterative application to an arbitrary initial image produces a definite convergence image as a good approximation of the original image. However, the image region with weak self-transformability would degrade the performance of the IFS image coding considerably. In this paper, we introduce a spatial image partitioning method based on a distortion-rate cost function, and then we study the following two improvement techniques : the extension to three-stage spatial image partitioning with range blocks of four different sizes, and the hybridization of the IFS image coding with a more traditional transform image coding.
