An Intelligent Restoration Method for Impulse Noise Highly-Corrupted Images

Abstract

The paper is dedicated to the restoration of impulse noise highly-corrupted images by exploiting the characteristics of the local similarity and connectivity existed in most real-world images. The basic strategy of the proposed method is firstly to detect a noisy pixel and then restores the corrupted pixel, by the local features of similarity and connectivity in an image. A decision rule based on the number of similar and connective pixels, followed by a line-judgement procedure, is used to determine if it is a noise. A simple local-connectivity (decision-based median) filter based on the noise density level is designed to restore the noisy pixel. Experimental results show that the proposed noise reduction method can remove impulse noise better than other methods in highly corrupted images of noise ratio more than 15%