摘要： 提出了一种小波系数的高斯混合分布模型。该模型用两个分量的高斯混合分布来拟合小波域每个高频频带系数的分布:其中,先验概率小方差大的分布代表了少量幅值较大的小波系数;先验概率大方差小的分布代表了大量幅值较小的小波系数。该文采用基于特征的快速算法完成对高斯混合分布参数的求解。实验结果证明了模型的有效性。

关键词: 小波变换, 高斯混合分布, EM算法

Abstract: In this paper, a Gaussian mixture distribution model of wavelet coefficients is proposed. We use a two-state Gaussian mixture to fit the distribution of the coefficients in the high frequency subbands after wavelet transform. The distribution with small prior probability and large variance represents a small amount of wavelet coefficients with large amplitude; and the distribution with large prior probability and small variance represents a large amount of wavelet coefficients with small amplitude. We solve the parameters of the Gaussian mixture distribution by a fast EM algorithm based on histogram feature. The experimental results show the validity of this model. ;;

中图分类号: 

• TN911.73