嵌入分层聚类的小波零树图像编码

Embedded Layered Clusters Wavelet Zerotree Image Coding

该文研究了小波系数的一类新的统计特性——分层聚类特性.为了利用这种特性,文章分析了这些分层聚类的结构关系,并结合已有小波零树技术提出了一个嵌入分层聚类零树小波图像压缩编码算法.这个新算法利用形态学算子来聚类优化子带内重要系数编码,同时使用小波零树来实现对子带间小波系数相似衰减特性的利用,提高对不重要小波系数的编码效率.因此,新算法有效地克服了单纯使用零树或单纯使用形态学聚类技术所存在的不足.实验结果证实:该文所提出的编码器性能相对于EZW编码器有明显的提高,其性能可以与SPIHT编码器和MRWD编码器相媲美.该文所提出的编码器不仅具有渐进传输和CBR应用所需的嵌入特性,同时也具有极低的算法复杂度.稍加改动还可以支持多分辨编/解码.

Theoretical analysis shows that a down-layer cluster always wraps an upper-layer one that having same seed when its structure element no less than one on upper layer. Therefore, by selecting a group of structure elements on successive bit plane layer, a progressive morphological representation for significant coefficients is allowed. Based on this observation, a layered cluster morphological representation is extended to the wavelet zerotree algorithm, namely the embedded layered cluster wavelet zerotree (ELCWZT) coding algorithm. The ELCWZT can be expected to overcome effectively the shortcomings that exist when using only zerotree data structure or morphological cluster representation. Experimental results on the standard test images show that the proposed ELCWZT algorithm outperforms EZW codec. It provides the performance comparable to that of SPIHT codec as well as MRWD codec in PSNR. The proposed algorithm also has many desirable features, such as the full embeddedness for progressive transmission and CBR traffic, and low-complexity. In addition to the rate scalability, its modified version also allows multiresolution scalability in encoding and decoding.