图像多小波域矢量量化策略

Vector quantization scheme in image multiwavelet domain

提出一种基于多小波变换结合矢量量化的图像编码算法(MDWT+VQ)。首先对图像进行多小波分解,然后对高频系数用改进后的LBG算法形成的码书进行VQ编码。算法充分利用了多小波域不同分辨率层间各方向子图像的相似性,仅对最高分辨率层进行码书地址索引,低级分辨率层的系数按照一定的组织形式直接套用最高分辨率层的地址索引信息。对比实验的结果验证了该算法在提高图像的重建质量以及在降低位码率方面均比传统的单小波图像编码算法有一定的提高。

An image coding scheme based on multiwavelet transform and vector quantization is proposed. Firstly, image is decomposed by multiwavelet transform, and then the high frequency coefficients are coded by use of the codebook made by the improved LBG algorithm. The scheme fully uses the comparability of different resolution sub-images in the multiwavelet domain. Coefficients in the highest resolution sub-image are indexed with codebook, and coefficients in the lower resolution sub-image use these index information directly under some certain organization. Comparing to the traditional coding method based on single wavelet, the scheme can improve the quality of reconstruction image and reduce the bite rate at the same time.