Research On Optical Image Denoising Based On Wavelet And Contourlet Transform

Optical images are very important for people to obtain information, but in the process of obtaining, transmission and storage, a great deal of noise is involved in and the quality goes bad. So image denoising is an important content in the digital signal processing. We do research on the denoising algorithms in the wavelet and contourlet domain detailedly and propose some improved algorithms. The paper is mainly organized as follows:Firstly, we introduce the research situation of wavelet denoising methods and point out the advantages and disadvantages. Then we introduce some fundamental knowledge, including wavelet transform theory, the characteristics of noise and denoising evaluation standard. These provide theoretical basis for the subsequent improved algorithm.Secondly, wavelet thresholding method is introduced. We conclude the advantages and disadvantages of the traditional thresholding function and then propose a new kind of thresholding function. In order to compensate the Gibbs phenomenon, we apply cycle spinning algorithm in the denoising method. The result of the experiment shows the excellent performance of the proposed method both in PSNR and the visual effect.Thirdly, we analyse the distributions of the wavelet coefficients of image and use Normal Inverse Guassian Model as prior model and Bayesian maximum a posteriori (MAP) estimator is used to estimate the noisy wavelet coefficients. In order to improve the behavior of Bayesian estimation, wavelet coefficients with different correlation are calculated with different ways. What's more, Cycle Spinning algorithm is used to modify the Gibbs phenomenon. Denoised images obtained with this method can remain edges and details well and has a high peak signal-to-noise ratio value.In the end, in order to represent the edges and contours more sparsely, we apply contourlet transform om the noisy image. We analyse the distributions of the contourlet coefficients of image and use Normal Inverse Guassian Model as prior model and Bayesian maximum a posteriori (MAP) estimator is used to estimate the noisy contourlet coefficients. At the same time, Cycle Spinning algorithm is used to modify the Gibbs phenomenon. Experiment results prove that the new method can remove Gaussian white noise effectively, reserve image edges better and enhance the peak signal-to-noise ratio.