A de-noising method based on connectivity strength between two adjacent pixels

Abstract

The essential idea of de-noising is referring to neighboring pixels of a center pixel to be updated. Conventional adaptive de-nosing filters use local statistics, i.e., mean and variance, of neighboring pixels including a center pixel. The drawback of adaptive de-nosing filters is that their performance becomes low when edges are contained in neighboring pixels, while anisotropic diffusion de-noising filters remove adaptively noises and preserve edges considering intensity difference between neighboring pixel and a center pixel. The anisotropic diffusion de-noising filters, however, use only intensity difference between neighboring pixels and a center pixel, i.e., local statistics of neighboring pixels and a center pixel are not considered. We propose a new connectivity function of two adjacent pixels using statistics of neighboring pixels and apply connectivity function to diffusion coefficient. Experimental results using an aerial image corrupted by uniform and Gaussian noises showed that the proposed algorithm removed more efficiently noises than conventional diffusion filter and median filter.