摘要： 对DCT域的分形编码算法进行了研究，提出了改进的算法。基于人眼视觉系统选择平坦块，选择出DCT域中的均匀部分，直接将其直流部分编码输出，不需分形匹配。为了降低匹配时间，在DCT域中，定义域块的8种变换的计算可以简化为两组内积。结果表明，在相同的匹配误差的情况下，该算法得到了更好的恢复效果，并且缩短了编码时间。

关键词: 分形, 离散余弦变换, 人眼视觉系统, 均方误差

Abstract: On the research of the fractal image coding in DCT domain, this paper presents an improved algorithm. In the improved algorithm, plain blocks are selected based on human visual system. Then by selecting out the uniform parts in the DCT domain, the direct current encodes directly without any fractal matching. For matching time reducing, the mean squared error computations of the eight orientations of the domain blocks are simply reduced into two groups of inner products in the DCT domain. The result shows for the same matching errors, this algorithm not only makes the image compressed effectively with high PSNR and much better visual effect, but also reduces processing time in an efficient way.

Key words: Fractal, DCT, Human visual system(HVS), Mean squared error(MSE)