Image Denoising Using Edge-preserving Filters

As an effective preprocessing method, image filtering is widely used in image engineering. An important goal of image filtering study is to present the image filtering method with structure preservation. Traditional filtering algorithms, such as wavelet filtering, anisotropic diffusion filter and non-local mean filtering, can acquire certain denoising effect for the contaminated image. But there still exists some deficiencies on noise suppression, structural retention and the algorithm's running time. The comprehensive result of the filter needs to be improved further more.The bilateral filtering algorithm can reduce the loss of the structure of the image in the denoising result, and the running time of the algorithm is shorter than most of the algorithms mentioned above. In addition, the implementation process for bilateral filtering is very easy. So it has been widely used in many image processing projects. The traditional (pixel-based) bilateral filter suffers difficulties in an appropriate selection of the smoothing parameter due to the disturbance of noise, which poses obstacles to the filtering process in dealing with edges and hence degrades the preservation of image structures. Meanwhile, the peak signal to noise ratio (PSNR) of the denoised image needs to be improved.For the effect of parameters'setting of bilateral filter to denoising result, this dissertation presents a novel algorithm with structure preservation that combines region segmentation and bilateral filtering to remove Gaussian noise in images. The process of bilateral filtering is directed by the region map which generates from the image segmentation.The innovation and the main content of this paper are as follows:1. The traditional filtering methods and their characteristics are summarized in this dissertation. Then several classical structure preserving-based filtering methods are reviewed and their advantages and disadvantages are compared and analyzed. The relationship among anisotropic diffusion filter, non-local mean filtering and bilateral filtering are described emphatically. And their relative advantages and disadvantages are also pointed out.2. This paper proposes a new algorithm for local statistical feature's calculation in images. Based on the segmentation of images, the algorithm takes advantage of the characteristics of the region map to facilitate the filtering process. The algorithm based on the region statistical feature can construct the adaptive bilateral filter by the variance of local noise to improve the denoising result.3. A new algorithm for denoising in images which is based on the difference of image's grey features between adjacent areas is presented. The algorithm use the the square of the average gray value between adjacent areas to weight Euclidean distance. The filter with refined effect of adaptability is realized by combining the statistical characteristics in the region map and the the difference of image's grey features between adjacent areas.4. A new algorithm for removing noise in images which combines the attribute on the regional boundaries is presented. The denoising algorithm between pixels on the regional boundaries is to propose a similarity function based on region boundary pixel's gradient value and the length of the regional shared boundary. The algorithm based on region similarity enhances the adaptability to image structures, improving the performance on the preservation of structure information.In this paper, four factors are adopted to evaluate the experimental results, PSNR, EPI, difference images and running time of the algorithms. The experimental comparison of a large number of real images with noise perturbation of various degrees and the experimental results relevant demonstrates that the algorithm in this work weights the denoising effect and edge preservation and the comprehensive effect of filtering is improved.