图像编码中的重要区域相关性研究及应用

在图像处理中，小波变换所具有的空频局部性，使得图像的能量重新分布。从函数正交分解的角度，小波分解相当于把图像分成一组正交基的坐标。由于这些坐标具有按不同频带所表现出的能量集中的性能，所以关于形态学的一些编码方法在基于小波分析的图像压缩中得到了很好的应用。因此，可以说小波变换在图像中的成功应用，主要归功于关于数据即小波系数系数和表示的重要性的认识。作者在文中提出了基于小波系数有效组织和表示的重要区域相关性编码（SRC），合理利用各子带间重要小波系数的相关性，通过聚类生成与聚类编码，在良好视觉效果与较高PSNR下，实现了图像较大倍数的压缩。

Research and Application for Significant Region Correlation of Image Coding

In image processing, energy re-distributes and reveals the localspatial and frequency property after wavelet transform. From the view point of function orthogonal decomposition, wavelet transform can be seen as translating an image into coordinates of a group of orthogonal basis. Every coordinate denotes the magnitude of energy according to correspondence basis. Because these coordinates have the property of energy concentration according to different subbands, the encoding method based on morphologic is well used in image compression. So the success in wavelet image coding is mainly attributed to the recognition of the importance of data organization and representation. In our paper, we provide a new method——significant region correlation coding (SRC) based on organizing and representing data efficiently. It is explicit to exploit the correlation of wavelet coefficient within all subbands by cluster generating and coding. We get a high compression rate as well as good visual perception and high PSNR. The SRC arithmetic mainly includes three steps. First, 7/9 Daubechies biorthogonal spine filter pair, which has the properties of linear phase and symmetric, is used to do the wavelet transform. After this step, image energy is distributed according to its fluctuating speed. Secondly, for different subbands, we use different thresholds and quantification times based on energy statistics and analysis. After quantification, we get almost 8 symbol maps per subbands consisting of Z, P and N. Finally, we organize these symbols by the method of morphology——cluster encoding. Before cluster encoding, too many small clusters should be removed for the little visual effect they caused. After cluster encoding, we use the method of significant symbol's address tagging and entropy encoding to finish the compression processing.