Bandelet Based Image Compression

With the development of Multimedia, the image technology attaches importance to everyone. Image science has the quality of rich in content. This dissertation is concerned with image compression principle, we has studied relevance theory, such as wavelet theory, wavelet alternation and Bandelet theory. Based on carry out the second generation Bandelet, we improved its capability. We can get much better effect in image compression. The main body of this article summary as follows:(1) We studied the characteristic from the first generation Bandelet to the second generation Bandelet, and mainly analyzed the basic theories frame of the second generation Bandelet, in order to assurance the Bandelet improvement direction.(2) We have implemented a Bandelet-based SPIHT coding method in this dissertation. This coding method divides Bandelet coefficients into different space direction tree according to level information. It makes use of tree structure to keep the relativity of the Bandelet inner part coefficients. The experiments show that the method can control the image compression rate conveniently, and give very good image compression result.(3) We put forward a compression method based on Bandelet and Support Vector Machine (SVM). This method mainly applied SVM regression analysis. The essence for the SVM regression analysis is inside error margin scope, using a handful of support vectors to make sure the superior regression flat surface. It can sparsely mean the original data gathers. Therefore our method can approach Bandelet coefficient by a handful of support vectors. When samples were trained, our method reserves important coefficient (support vector) only, and reduces around a great deal of unimportant coefficient in order to attain the purpose of data compression. The experiments show that our method is more obvious advantage at higher compression rate.