Wavelet Thresholding Denoising Alogrithm Research Based On The Sure Theory

Noise is an important factor affecting the image quality, the exist of the noise can lead to the unrecognizaton of some detail characters and decreasing of the peak signal to noise, therefore how to reduce noise effectively and extract image information becomes very important. In recent years, denoising methods based on the wavelet transform has been more and more widely studied and applied. The image wavelet denoising first consructs the corresponding rules according to the different representation of image signal and noise in wavelet domain, then processes the image signal cofficients and noise cofficients. The essence of processing lies in reduce the noise cofficients as much as possible even or eliminate the noise cofficients completely, while maximumly retain the cofficients corrsponding to the effective image signal.There are many denoising methods based on wavelet transform, this paper mainly discusses the thresholding denoising method which is a one way of denoising methods. Firstly the principle of thresholding denoising is generally introduced and the choice of some key parameters in the denoising process is generally analysed and explained, for example the choice of the thresholding and thresholding function. then, some choosing principles are given. Secondly, some typical denoising methods in the wavelet denoising methods are introduced, including basic thresholding function and thresholding based on the statistical model. then simulation experiments and comparison are done.The thresholding denoising methods based on statistacal model have good performance , especailly BLS-GSM model. It is better than other methods because of considering four characters of the cofficient: sparity, transition, aggragateion, orientation. However, this method needs to make a hypothesis of standard probability distribution model for the image signal, so it is very complex and needs large calculation. Therefore ultimately a improved thresholding denoising algorithm based on the linear square estimate theory is proposed. This algorithm do the linear expansion for a basic thresholding. According to the simulation experiments of gray image and RGB image, it can be proved that this proposed method have good denoising effect.