Research On Image Denoising Based On Wavelet Transform

Noise will disturbe images by using all kinds of methods in image transporting and receiving. It should be denoised when transporting and processing in order to improve quality and facilitate the successive higher level. Effective protection of the edges or textures is in contradiction to removing the noise in the process of denoising. Therefore, researchers make it a goal that finding a method which can remove noise effectively while well protects the edges and textures of lmage.Wavelet transform can express the feature by using time domain and frequency domain. It has well partial specific property and can do partial image analysis in time and frequency domain. So, as a tool of image process, wavelet transform get wide application. This paper based on the wavelet transforms theory, research removing of gauss noise in image. Its main works are as following:Firstly, it introduced denoising by using several basic wavelets transform and elaborate the basic principles and algorithms. As to wavelet thresholding, we focused on two key issues that threshold determining and choosing of threshold function. It introduced commonly-used threshold determine methods and threshold functions, finally compared the simulation results.This paper is based on the BayesShrink which SG Chang, etc proposed. It uses this method to determine different thresholds for each subbands and orientations. I proposed two methods based on multi-threshold. The first one is set a rate for each threshold, then use single wavelet to de-noise. The second method is using an improved threshold function. This function can overcome non-continuity at its threshold absolute value of hard threshold and the constant bias between soft threshold and original image through modifying a parameter. This method uses multiple wavelets to remove noise from images. I make simulations for the classical methods and improved methods. The results shows that the second method get better effect when compared with classical methods and the first method.