Abstract: A novel method of building prediction model based on local entropy minimum is proposed, which can be utilized to improve coding performance of wavelet coefficients. Firstly, prediction coefficients are selected according to the correlations between wavelet coefficients. Based on this, several prediction functions are constructed by combining prediction effects of the multiple coefficients. Secondly, a method of building prediction model based on local entropy minimum rule is proposed, which selects functions according to their prediction effect step by step. Finally, a wavelet prediction coding algorithm is designed based on entropy coding. The experimental results show that, the reconstructed image objective quality（PSNR） of the proposed prediction coding method is averagely 0.4 dB higher than image compression standard JPEG2000.

Key words: entropy, prediction coding, wavelet coefficient, image compression

摘要： 提出了一种基于局部最小熵的预测模型构造方法，能够更好地区分待编码位的不同概率分布，从而实现对小波系数的高效压缩。首先，根据小波系数间的相关性选择预测系数，并构造相关性预测函数来综合多个系数的预测效果；以熵值的最小化作为准则，采用逐步筛选法对预测函数划分的多个分类进行选择合并，建立了一种局部最优的预测分类模型；结合熵编码实现对小波系数的高效压缩。实验结果表明，与图像压缩标准JPEG2000相比，所提方法的恢复图像主客观质量均有改善，客观质量平均提高0.4 dB。

关键词: 熵, 预测编码, 小波系数, 图像压缩