Abstract: The search method based on classification in accordance with average gray value，which uses MMSE（Minimal Mean Squared Error），can better improve the time expenditures of fractal image coding procedures，but limit the quality of the image reconstructed after decoding.In this paper，to obtain better image quality，a new error rule which makes use of variance’s reflecting changes of image gray level is proposed.The optimal match results of traditional way that use MMSE are improved again.Simulations show that after adding the new rule，higher PSNR of decoding image is achieved with no affection on time expenditures.

Key words: fractal image coding, Minimal Mean Squared Error（MMSE）, optimal match, variance, Peak Signal to Noise Ratio（PSNR）

摘要： 基于传统的最小均方误差准则（MMSE）的灰度均值分类搜索方法，在有效改善分形图像压缩编码时间的同时使图像质量受到了限制。为了进一步提高图像质量，该文利用方差反映图像灰度变化这一特征，提出一种新的基于方差的误差准则，在MMSE准则的基础上，对编码得到的最佳匹配结果再进行改进。实验表明，在编码中引入该准则，解码后的图像拥有更高的峰值信噪比，而且编码时间并没有受到影响。

关键词: 分形图像编码, 最小均方误差, 最佳匹配, 方差, 峰值信噪比