Study On Wavelets And Support Vector Machines And Their Application In Image Compression

Efficient encoding algorithm is the key for image storage and transmission. Original image data needs huge storage space, and is unfavorable to store and transmit.To decrease the image storage space and realize real-time data process, it needs a high-perfomance image compression algorithm. Image compression code is divided into loseless compression and lossy compression.The classical image compression algorithm has already been sucessfully developed. Lately, some new compression method is deeply researched, such as image code based on wavelet transform, fractal theory and neural networks, which is called as the second-generation image code method, and are becoming increasingly concerned as a result of its higher-quality image compression effect. But because of their faulty theory, there are still many defects in factual application. According to research states of image compression,we deeply research image compression algorithm based on support vector regression and wavelet transform, and the main works are described as following.(1) We deeply research an image compression algorithm of combining support vector regression and discrete cosine transform, and experimental results show its validity.(2) Combining SPIHT structure, we propose a gray image compression algorithm based on SVR and wavelet transform. First, Original image data is decomposed by wavelet transform to gain multiresolution image data. Secondly, wavelet coefficients are resorted by using the SPIHT structure, and then we use support vector regression to process these wavelet coefficients on each SPIHT to obtain a series of support vectors and their corresponding weight value. At last Less support vectors are used to fit the primitive wavelet coefficients, thus the target of data compression is achieved.(3) Combining 3DSPIHT structure, we research a multi-spectral remote-sensing image compression algorithm of combing SVR and wavelet transform. Based on image compression algorithm with two-dimensional SPIHT coefficient trees, we extend it to image compression algorithm with 3DSPIHT. The multi-spectral remote-sensing image are decomposed by two-dimensional wavelet transform, and then their wavelet coefficients are reorganized and resorted by 3DSPIHT structure, and lastly are compressed by support vector regression. Experimental results show this method may gain preferable effects. But in Data Compression, longer processing time of support vector regression and image compression need be solved in future research.