Cluster based Non-Local Filters for Colour Image denoising

Abstract

Non-local filters have evolved as an alternative to local filters for image denoising. Even though these filters relay on the similarity of pixel neighbourhoods within a search window, there are chances that similar pixel neighbourhoods may not be present in the search window. This problem is resolved in cluster-based NLM (CNLM) - a recent modification to the basic NLM. CNLM suggests an alternative method for the formation of search window. The pixels in a given image are divided into clusters. Clustering is carried out based on the similarity of neighbourhoods of pixels. A given pixel P belonging to a cluster C is modified as the weighted sum of all other pixels in C. In this paper, we have carried out an extensive study of the CNLM algorithm using colour images. Further, we experimented the cluster approach with non-local Euclidean median algorithm (NLEM) and improved non-local Euclidean median algorithm (INLEM). The study reveals that the performance of cluster-based approach is better than the respective basic algorithms for low noise densities in terms of performance evaluation measures.