Research On Image Processing Algorithms Based On M-Band Wavelet

Wavelet Transform has been widely used in image denoising, compression and index for its multiresolution analysis. Currently the researches depend mainly on traditional 2-Band wavelet. But, the 2-Band wavelets can only decompose signal into two parts, the low-pass band and high-pass band. The bandwidth of high-pass is wider than its in low-pass, so its analysis results are better for the signal that the energy of it is more concentrated in low-pass or high-pass band. But for the signal that the energy is more concentrated in middle bands or distributed in several bands, the results of decomposition are not optimal. The base of 2-Band wavelets can not be orthogonal (energy concentration) and symmetric (linear phase) simultaneously.To overcome these shortcomings, many people have proposed that using M-Band wavelet transform for more finner analysis.The M-Band wavelet transform can decompose the energy of a signal into several band-pass subbands that still have characteristic of energy concentration and linear phase. Because of these characteristics, the M-Band wavelet transform has better performance than 2-Band wavelets in image processing.In this thesis, based on the research of threshold original from generalized gaussian distribution, an adaptive soft threshold that can be used in M-Band wavelet transform is proposed. We found that threshold of the subbands in different scale at multiscale decomposition is not only related to subband at different scale, but also related to the size of the subband. The experimental results show that the new threshold algorithm has very good performance, especially for the noised images that have more contextual information.With the development of multimedia and internet, the image libraries become larger and larger. It has become the key technology to search for exact image information from the database rapidly. It is also named image index technology. The CBIR technology is becoming a very active research field. The M-Band wavelet transform can decompose the high frequency signal into more meticulous subband, so it must have better performance in image index. In this thesis, a CBIR system that is based on M-Band wavelet transform is used which use the Euclidean distance as the measure for similarity and adapted five normal textual characteristics consisted of energy, and logarithm et al. The experiments use Brodatz image library as test object to perform index for whole library and training. The results show that the CBIR system based on M-Band wavelet transform has better performance.