Optimal Fuzzy System Design for Additive Noise Removal in Color Images

Abstract

Image Denoising is a fundamental preprocessing task before further operations like image segmentation, feature extraction and texture analysis etc. of the image can be carried out. The purpose of denoising is to remove the noise while retaining the edges and other detailed features as much as possible. In this work, a filter to remove additive noise i.e. a combination of Gaussian noise and Salt & Pepper noise in color images is presented. Image Denoising is performed in two steps. First, an edge map is calculated on the noisy image using a modification of bacterial foraging. Bacteria’s will move on to the edge pixels thus calculating an edge map. The criterion for a pixel to be chosen as an edge pixel is described. Bacterium health is also maintained in terms of edge pixels found by that bacterium. In second step, we will apply Denoising on image pixels leaving edge pixel as it is. In Denoising technique, pixel based similarity concept is used. Pixel similarity is defined in fuzzy way since similarity is an imprecise concept. Fuzzy peer groups are constructed. Every pixel in the surrounding of the pixel in interest will belong to the peer group set of that pixel with some membership value. Membership function is defined in such a way that each member will belong to the set with its membership value between 0 and 1. Then a best no. of members is chosen among all members that are in the peer group depending on some fuzzy rules. Then we will use this fuzzy peer group to filter the noise in two cascading steps. First, an impulse noise filter is defined depending on fuzzy rules. If a pixel is detected to be corrupted by salt & pepper noise then it is corrected by taking vector median filtering of fuzzy peer group pixels. Then Gaussian noise is filtered by taking weighted averaging of fuzzy peer group members with its membership value to the fuzzy peer group. Experimental results showed that discussed technique is able to calculate edge map with high entropy value as compared to other edge detection methods. The discussed filter is able to remove a mixture of both salt and pepper noise and Gaussian noise with promising results. Edges, image details are also better preserved.