Research Of Methods In Image Denoising Based On Contourlet

Image denoising is an important part of image processing, which directly affects the results of the following treatment, but the traditional denoising methods tend to smooth the noise, while losing the detail information of image. Because the multi-scale geometric analysis has the advantage of multiresolution and multidirection, it has been widely used in the field of image denoising recently, and as one member of it, Contourlet could approximate the best representation of image, this paper studies of the image denoising application of Contourlet and TICT(Translation-Invariant Contourlet Transform) deeply. Works as follows:1.Combining Contourlet with a self-adaptive and multiscale analysis method: Treelet, according to the multiscale decorrelation feature of Treelet, apply it to the decomposition and dimensionality-reduction of the high-frequency coefficients in Contourlet domain, it can analyze coeffients’ potential structure and correlation between each other to estimate a self-adaptive threshold, propose a image denoising method based on Treelet in Contourlet domain. This scheme maintains image’s details while eliminating pseudo-Gibbs phenomenon up to a certain extent, it also makes the image more clear.2.Proposing a denoising method of Wiener filtering in Contourlet domain based on2DOtsu. Signal variance estimate is very important in Wiener filter, we can get perfect result if the variance being estimated accurate.So we first get the important coeffients which represent signal through separate high-frequency coefficients by2DOtsu. then select local neighborhood in different direction and different scale estimate signal variance and filter. The experimental results indicate the estimates signal variance is much accurate, the method is also proved simple and efficient.3.When Contourlet decomposes image, the bandpass image decomposed by Laplace pyramid filter and DFB filer in Contourlet will produce oscillation and aliasing phenomenon around singularity. So we further research the TICT, according to the heavy-tail feature of TICT coefficients, generalized laplace distribution is used to model the noise-free coefficients while gussian distribution is used to model the noisy coefficients. Through iterative operation we get the probability of noise-free coefficients in all coefficients, then denoise by mixed distribution filtering. Experiment indicates this method can smooth the noise of image effectively and it gets a high PSNR.