Research On Image Denoising Method Based On Contourlet Transform

Image denoising is a very important and basic research in the field of image processing,whose effect has significant influence on subsequent processing of the image. As a very effective method of multiscale geometric analysis, Contourlet transform has very good properties such as multidirection and anisotropy, as well as the properties of multiresolution and time-frequency localization inherited from wavelet transform. So the image denoising algorithm based on contourlet transform has a good prospect of application.Based on the research of Contourlet transform and the existing image denoising methods,this thesis analyzes the coefficient distribution characteristics of the image after contourlet transform. According to the coefficients distributed in multiple directions and coefficients with relevant features within the subband, two new image denoising algorithms are put forward in this thesis. Main content of the work are summarized as follows:(1) Traditional threshold denoising method takes the same handling for every subband coefficients. At the same time, there are some defects in the traditional threshold function,which is easy to cause the loss of image information. So an improved threshold and improved threshold function in contourlet domain for image denoising are proposed. The improved threshold is the weight threshold based on particle swarm optimization(PSO). According to the distribution characteristics of contourlet coefficients of different subband, the weight threshold of different subband is obtained, and then PSO algorithm is introduced to optimize weight parameter for getting more optimal threshold. Taking the energy distribution of different subband into consideration, the threshold will be more reasonable and can reduce the shortcomings caused by the same handling to all coefficients. On the other hand, improved threshold function proposed in this thesis is expressed as piecewise continuous function,which overcomes the shortcomings of the traditional soft and hard threshold function, and handles the different coefficient with different processing to avoid image gibbs effect and fuzzy edge information.(2) Because the application of traditional wiener filter in transform domain is easy to cause image detail information fuzzy, the improved wiener filter in contourlet domain is put forward in this thesis, and combined with partial differential equation for image denoising.Due to the importance of contraction factor in wiener filter, this thesis firstly classifies the contourlet coefficients into different kinds and estimates the contraction factor based on the signal coefficients, then processes the contourlet coefficients with different wiener filtering.The improved wiener filter truns out to be effective in keeping edge information of the image.The total variation(TV) model based on partial differential equation is used for further denoising in this thesis, which not only makes a better denoising effect but also overcomes theshortcoming of over smoothing. The improvement of wiener filter and the using of TV denoising model effectively make up for the shortcomings of each algorithm, and achieve a satisfactory effect.