Research On Image Denoising Methods Based On Non-Subsampled Contourlet Transform And Bilateral Filtering

The image is one of the most important source of human to get information. with the rapid development of science and technology, image processing techniques is also developing rapidly. However, And image after image components inevitably will be affected by the interference, the image is noise pollution. In order to effectively for subsequent processing,in order to extract the information effectively from noisy image, we need the help of image processing technology to reduce the noise. Because of its good time-frequency wavelet transform and multiresolution features such as widely used in the field of image processing technology. But wavelet transform can only reflect the point singularity. non-subsampled Contourlet transform Multi-scale geometric analysis method is put forward, make up for the inadequacy of wavelet transform. This article mainly introduced the next sampling Contourlet transform, on this basis, put forward the improvement of the next sampling Contourlet transform image denoising algorithm, the main contents of thesis are as follows:(1) This paper expounds the research background and significance of the image denoising, multi-scale geometric analysis and classic denoising algorithm. The disadvantages of each Spatial filtering algorithm and Frequency domain filtering algorithm were summarized and analyzed. which provides the Theoretical basis for the research of image denoising algorithm under the non-subsampled Contourlet transform.(2)we have analyzed the flaws of the general threshold, a new threshold algorithm was proposed.in order to make up for the disadvantages of the hard threshold function, the soft threshold function and the half-threshold function, The new threshold function is continuous. By experimental verification, the improvement of non-subsampled Contourlet transform denoising is feasibility and the algorithm is effectiveness.(3) The main factors influencing the bilateral filter denoising effect filter width, Spatial neighborhood degree factor and pixel similarity factor, The strength of the noise affect the width of the filter, The traditional bilateral filtering is not adaptive to parameter selection, We propose an adaptive factor adjusting the filter width according to the noise intensity. Can adaptive filter out noise.(4) In the traditional threshold denoising algorithm, since most of the noise exists in the high frequency coefficient, so only to process the high frequency coefficients in transform domain, however still contains a small amount in the low frequency noise, but will not be processing,and direction subbands of each high-pass component is processed by the new threshold function which is obtained by the Bayes threshold that based on stratified noise estimation. During the reconstruction, the low-pass subband constructed image is further denoisied by the bilateral filtering in the spatial domain. Experimental results demonstrate that the proposed method can improve de-noising performance.