基于分形搜索树的嵌入式小波图像编码算法

Fractal-Searching-Tree-Based Embedded Wavelet Image Coding

与单纯采用分形编码方法相比,基于小波的分形图像编码可以较好地解决方块效应问题且能够有效降低匹配搜索时间,但在低频子带使用分形编码会导致重构图像质量下降,同时针对匹配搜索仍是分形编码主要时间开销的问题,提出一种基于分形搜索树的嵌入式小波图像编码算法.采用Haar小波对图像进行多级分解,对低频子带直接采用DPCM编码,高频部分则依据不同尺度子带的重要性采取自适应方式划分值域块,然后构建一种分形搜索树结构以确定定义域池并采用"Z"形扫描进行匹配搜索,最后对获得的分形参数进行算术编码.实验结果表明,该算法重构图像质量比同类算法有所提高,特别在中低码率下PSNR值提高明显,当码率小于0.40bpp时,PSNR平均提高0.40～2.48dB,同时算法执行时间明显减少.

Compared with fractal image coding, wavelet based fractal image coding can cope with block artifacts and reduce match-searching time effectively. But using fractal coding in low frequency sub- band will lead to poor reconstructed image, and the match-searching time is still the main overhead for fractal image coding. So a fractal-searching-tree-based embedded wavelet image coding algorithm is proposed. The image is decomposed to multiple-level sub-bands by means of Haar wavelet. For the low frequency sub-band, the DPCM coding is applied directly. For the high frequency sub-bands, a self-adaptive approach is adopted to partition each sub-band into different range blocks according to the significance of sub-bands with different size. Then a fractal searching tree structure is constructed to determine the domain pool in which match-searching is carried out in a manner of zigzag scanning. Finally, the arithmetic coding method is employed to encode the fractal parameters obtained. Experimental results show that better reconstructed images are obtained as compared with those by other similar algorithms, and the PSNR is remarkably improved in medium and low bit rate. When the bit rate is less than 0.40 bpp, the PSNR is promoted about 0.40-2.48 dB averagely. Meanwhile the running time of the algorithm is also reduced.