基于感知度量的分形编码

A Fratcal Image Coding Based on Visual perception

传统的分形编码利用最小均方差准则（ＭＭＳＥ）来度量图象子块的变换匹配，未能充分利用人眼的视觉特征。本文提出了基于人眼视觉模型的分形编码方案。算法中，按视觉上的差异对图象子块进行分类；由“视觉最相似”的准则确定最佳匹配的域块，为此，相似匹配转换到ＤＣＴ域进行，导出了图象块的相似匹配在ＤＣＴ域的形式；算法进一步通过自适应误差校正来消除解码图象中的方块效应。解码图象的视觉质量有了明显的改善。实验结果证实了算法的有效性。

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.