Fuzzy non-local image guided filter and averaging

Non-local model has been a very useful tool in image processing for its global information accumulation. As we all known, the patch centered a fix pixel in an image is in dynamic change when adapting some image processing algorithms, and in the classical non-local model, the weights are set to be constant. For example, in the non-local guided filter methods, the non-local weights need to fine tune according to other parameter and the guidance. To fix this problem, we improve the non-local models by employing the fuzzy sets theory. In this way, the global information are accumulated with fuzzy weights in its iterative optimization. In this paper, a fuzzy non-local image guided filter and average methods (FNLGFF and FNLGAF) are proposed and show good performance in many image processing tasks by utilizing the fuzzy weighted non-local similarity of the guidance image. This fuzzy model shows more reliable weighted filtered and averaged image results when handle image noise by update the non-local information in every iteration. In this process, it also suppresses the low similarity values of the guidance image and boosts high similarity values. Experimental results on several image processing tasks including image denoising and image dehazing verify the superiority of our fuzzy non-local image guided approaches.