Local Directional Denoising Methods Based On Multiscale Space And Its Applications

Image information is the main source that human gain information from the objective world. However, in the progress of the image's generation, coding even reconstruction, the image may be influenced by noise and become noised image. That how to denoise the image to improve the image quality becomes a very important task in image processing.For the past few years, the basic idea of Multiscale Geometric Analysis(MGA) has developed a series of new theories independently in many areas such as mathematical analysis, computer vision, pattern recognition, and statistical analysis. In the area of image processing, the goal of MGA is to detect, show, and deal with the data of the higher dimensional space. The key characteristic of the data is shown intensively in the low dimensional subsets.This dissertation focuses on the local directional denosing methods based on the multiscale space. In the multiscale space, according to the directional feature of the coefficients in different subbands, we estimate the coefficients' variance in the local neighborhoods which are rectangle windows of different sizes and directions, then denoise the image by the Wiener filtering . The experimental results show that the proposed algorithm improves the denoising performance significantly.The main work of this dissertation includes:1,Employ the wavelet transform and Wiener filtering, and propose the idea of local direction denoising. In wavelet domain, according to the features of wavelet transform, choose different directional rectangle windows as local neighborhoods in the different subbands . Then estimate the variance using the MAP method and denoise the image by the Wiener filtering. The experiment results show that this algorithm is not only low computational complexity and easy to implement, but also obtaining high PSNR and protecting the detail information of the image.2,Because of the directional limitation of the wavelet transform, we research in the dual-tree complex wavelet transform (DTCWT), and do experiments of its local direction denoising method. 3,For the wavelet transform has some defects, MGA(Multiscale Geometric Analysis) has been proposed and developed quickly in recent years. It has shown more advantages than the traditional methods in the application of image processing. The contourlet transform is one of the hot spot in the MGA. In this dissertation, we study the Contourlet transform and its application in image denoising.The dissertation includes following six parts: Section 1 shows the background and significance of this research. Section 2 introduces the development of wavelet transform and the threshold method. Simulation experiments are also implemented in this part. Section 3 is about the denoising method of rectangle windows based on the wavelet transform. In Section 4, DTCWT is introduced, and the local direction denoising method is carried out. Section 5 introduces the fundamental principles of contourlet transform. Contourlet transform can catch the marginal information of images accurately into different scales and frequencies subbands, so it is well used in image denoising. This part mainly studies the local neighborhood denoising method in contourlet domain. The dissertation ends with some further research plans in Section 6.