Partial Ellipse Filter for Maximizing Region Similarity for Noise Removal and Color Regulation

摘要

Ellipse filters can be implemented for partition of image patch. We introduce a method for automatically obtaining a set of neighbor patches for an image pixel in form of ellipses for noise removal and color regulation. Comparing neighborhood similarity of patches of the set allows selecting an optimal patch. The evaluation of similarity is developed from bilateral filter with additional orientation condition. Through the development of the image filter model it is shown that the image noise can be removed better with the ellipse patches that are allocated in different directions. Our first finding is that it is enough to select 4 or 8 major orientations to determine the best ellipse patch for each pixel. Secondly, by operating convolution weighted by intensity similarity and the spatial distance, this is capable to detect particular oriented patch with the best neighbor similarity and ameliorate the elimination of different noise types. These filters also permit remaining color harmony and edge contrast for color correction. In particular, the validity of the method is demonstrated by presenting experimental results on a benchmark database.