Research On Image Denoising Method Based On Wavelet Transform

Wavelet image denoising has been one of the main methods of image denoising. It has been a generally-studied subject to denoise by using the de-correlation and the multiresolution properties of wavelet, the statistical properties of wavelet coefficients and the dependency of inter-level and intra-level coefficients. In the image denoising algorithms, it is also a hotspot to overcome the lack of shift invariance of discrete orthonormal wavelet transform.In this thesis, wavelet image denoising methods have been analysed, and the shift-invariance wavelets have been discussed. First, a few classical wavelet denoising methods are introduced in detail, and the experimental data are presented. In the proportional shrinkage denoising method, we deduce the estimation of original image coefficients using the simple linear regression method. Furthermore, we propose another two bivariate shrink denoising models based on the BivaShrink method. The two models and the BivaShrink method use the mutual information of two of the current coefficients, the parent coefficients and the adjacent coefficients respectively. Taking into account the mutual information of the three, a trivariate shrinkage function (TrivaShrink method) is brought forward, which considers the inter-level and intra-level dependencies of wavelet coefficient.In addition, we analyse the theories and properties of complex wavelet in detail. The merit of both shift invariance and directional selectivity enables the complex wavelet to overcome the artifacts in the discrete orthonormal wavelet denoising. We make some denoising experiments by combining the complex wavelet with the two denoising models and TrivaShrink method. From the experiments, it is indicated that the complex wavelet has obviously better performance in both the denoising error and the image visual quality.