In classical fratcal image coding

the self-similarities between image blocks are measured by MMSE. No human visual properties are considered. In this paper

we propose a novel fractal block coding based on human visual system. The original image blocks are classified by their visual differences; For each range block

the domain block which is“the most similar in vision”is searched and the matching of blocks is completed in the DCT domain; Adaptive distortion equalization is used to reduce the blocking effect in the decoded image. The performance of our method has a visible improvement of subject quality depending on visual perception.