Research On Image Compression Algorithm Based On Wavelet Transforms

Digital image compression is a hot research subject of image processing, the achievements of which bring great convenience to the storage and transmission of images. Thus it is of significant value for researchs. Among various image compression methods, the method based upon wavelet transforms is of obvious advantages and therefore has become a major research subject of modern image compression.The dissertation analyzed several classical algorithms of wavelet image compression and the encoding methods of Region of Interest (ROI) in JPEG2000 image compression standards, both of which leave some flaws to be further improved. The main work of the dissertation is as follows:1. A brief description about the theoretic foundation of wavelet transforms was given, which includes continuous wavelet transforms, discrete wavelet transform, Multi-resolution analysis, Mallat algorithm, double orthogonal wavelet transform, lifting wavelet transform and integer wavelet transform. A comparison was made between several representative algorithms of wavelet image encoding, including two algorithms of tree structure: Embedded Zerotree Wavelet (EZW) and Set Partitioning In Hierarchical Trees (SPIHT), and two algorithms of block structure: Set Partitioned Embedded Block Coder (SPECK) and Embedded Block Coding with Optimized Truncation (EBCOT). Besides, the deficiency of the tree structure algorithm was analyzed.2. We proposed a hybrid encoding algorithm combining SPIHT algorithm with fractal encoding algorithm. It classifies the partitioned blocks of image into three categories, and uses SPIHT or fractal method in wavelet field according to different features of the categories, and makes use of rate distortion criterion to achieve better image quality than those of SPIHT and fractal methods.3. By analyzing the merits and demerits of the two ROI encoding methods of JPEG2000, i.e., Maxshift and Generic scaling based method, we proposed a Maxshift based algorithm using transition region, which meets the need of human vision features and gets better decoded image in terms of overall visual effect. Moreover, the proposed algorithm is of low computational complexity and is compatible with JPEG2000 code stream.