Weighted Linear Integral Convolution Type Image Denoising

The study on the image denoising method is not only significant in practical applications, but also important for theoretical study of image processing. Since image denosing is very closely associated with image regularization theory and modeling theory. This paper mainly focuses on the research of partial difference equation based image denosing methods. Starting from the review the development history of PDE based image denoising models, we focus on the analysis of three popular diffusion equation models, such as tensor type div-based PDE model, tensor type trace-based PDE model and curvature-preserving PDE model. The pros and cons of the models mentioned above are pointed out, and we get insight into how to improve the denoising performance of the anisotropic PDE modes. Based on this, we investigated curvature-preserving PDE model, and the main results have been proposed as following:1) A weighted linear integral Convolutions based image denosing method is proposed. Firstly, we employ total variation flow technology to smooth the structure tensor, and yield nonlinear structure sensor which can be construct the diffusion tensor; Then, the diffusion tensor is projected into different directions, and produce corresponding vector fields, by which we can compute each integral convolutions results; Lastly, the weighted function is designed by directional information. Experimental results show that our new method can improve the performance of the curvature-preserving PDE model.2) A patch based weightd linear integral Convolutions image denosing method is investigated. This method is an improved version of the weighted linear integral Convolutions based denosing method, we take advantage of image patch processing technology in nonlocal means filtering, which replace Gaussian function in linear integral Convolutions operation with the image patch similarity functionExperimental result shows that the improved method have superior performance on preserving image edge and curvature geometric structure.