| Images are usually contaminated by various kinds of noises during the gathering and transmission process. Noises will bring degradation to the content of source images, thus making the following image processing tasks like image segmentation, target recognition, image retrieval, image coding and transmission more difficult to handle. Therefore, image denoising play a key part in most image preprocessing applications. The idea of image denoising algorithms is to eliminate noise and remain the original image information as much as possible, that is, image denoising problem can be treated as nonliner approximation of the original image.The classical image denoising algorithms in spatial domain and transform domain are analyzed firstly, and their advantages and disadvantages are also discussed. In this section, the focus is put on the image denoising algorithms based on wavelet transform. In recent years, a 2-D signal processing theory called multiscale geometric analysis (MGA) has been emerging, which can effectively overcome the disadvantages of conventional wavelet analysis. Based on previous researches, a novel mage denoising algorithm is develped combining multi-thresholding and a MGA tool-contourlet transform. Contourlet was constructed to achieve better multresolution, mutidirection, anisotropy and sparse properties, so that better image processing performance can be obtained in comparison with wavelet transform. The author discusses the construction of discrete contourlet transform and its extension to continuous domain, and proves the validity of image processing via contourlet transform. The proposed contourlet based denoising algorithm adopts the improved hard thresholding function with an optimal thresholding set. The distribution rule of the transform coefficients and the statistical characteristics are used in the thresholding function. The optimal threshold set is adaptively determined by the number of decomposition scale and the intension of the noise. Experiment results show that compared with wavelet denoising methods, the proposed algorithm can achieve great improvement in both visual effect and objective quality. |
