图像矢量量化的等误差自组织特征映射算法

Vector quantization based on equidistortion self-organizingfeature mapping algorithm

提出了一种使各区域子误差相等的矢量量化算法,算法利用小波变换后各子带间的相关性,合理构造矢量。采用最优矢量量化器设计原则,使用改进的自组织特征映射算法生成码书。学习过程中,通过调整各子区域的误差,使之趋于相等,改善总的期望误差,获得更接近全局最优的码书。实验表明,这种算法获得的码本优于其它几种算法。

A vector quantization algorithm with equidistortion on each region is introduced. The correlation of the wavelet coefficients in different frequency bands is effectively exploited through constructing the band-cross vector. Using the designing principle of the best vector quantizer, an improved Self-Organizing Feature Mapping (SOFM) algorithm is used in constructuring codebook. A nearly optimal codebook can be obtained by modulating the subdistortions of all regions so that they are equal to each other. Experimental results show that the algorithm is superior to other methods.