Study On Image Denoising And Texture Classification Based On Wavelet Theory

Wavelet analysis has become a powerful tool of transient signal processing and has been widely used in the field of image processing. image denoising based on wavelet-domain statistical model is more effective, and it is recently the hot field of research. This dissertation focuses on the statistical model of wavelet coefficients and its application in image denoising and texture analysis. The main work can be summarized as follows(1) we discuss image denoising methods based on wavelet maxima, wavelet thresholding and Bayesian estimation. A detailed analysis is given to threshold and thresholding function selection . we paid more attention to the Bayesian model of wavelet-based image denoising, and give a Bayes estimate of wavelet coefficient under the three most commonly used loss functions.(2) give a probabilistic model for an individual wavelet coefficient. We model each wavelet coefficient as a random variable with the BKF distribution with two unknown parameter. We have shown that the BKF distribution is very powerful in capturing the heavily-tailed behavior of the wavelet coefficients over a large class of images. The BKF model also demonstrated a high degree of match between observed and estimated prior densities of wavelet coefficients. A Bayesian denoiser with the BKF prior is proposed, and experimental results show that the peak signal-to-noise ratio and the visual quality of the proposed denoiser is superior to the other methods.(3) the hidden Markov tree models use a probabilistic tree to model Markovian dependencies between the hidden states to capture the dependencies between the wavelet coefficients, and it is useful tool for image processing. Unfortunately, models based on the orthnal wavelet transform suffer from shift-variance, making them less accurate and realistic. In this chapter, we extend the HMT modeling framework to the complex wavelet transform. Finally we propose a new wavelet statistical model, called bivariate BKF probability model, which able to better capture the interscale dependencies of wavelet coefficients.(4) first the dual-tree complex wavelet transform(DT-CWT) is discussed, The DT-CWT not only inherits the advantages of tradition wavelet transform, but also possesses nearly shift invariant, directionally selective, limited redundancy and efficiency calculations.Image denoising based on the DT-CWT can eliminate the Gibbs phenomenon. Finally two image denoising algorithm in the DT-CWT domain is presented. In a simple estimation experiment, the denoising algorithm based bivariate BKF model outperforms a number of high-performance denoising algorithms.(5) a texture classification algorithm in the DT-CWT domain is presented, nearly shift invariance and good directional selectivity properties of DT-CWT make it a good candidate for representing the texture features. 30 images in Brodatz texture datebabase are well classified with the presented algorithm.