Cluster Based Non-Local Approaches for MR Image de-noising

Abstract

Non Local Mean filters (NLM) are an efficient alternative to local window based de-noising algorithms. NLM filters are found to be suitable for enhancing the quality of medical images, especially those degraded with Rician noise. In this paper, a cluster based non local mean filtering algorithm is proposed. The cluster based approach is incorporated into three major versions of NLM, namely basic NLM, Non Local Euclidean Median algorithm (NLEM) and Improved Non local Euclidean Median algorithm (INLEM). This resulted in three cluster based algorithms – Cluster based NLM (CNLM), Cluster based NLEM (CNLEM) and cluster based INLEM (CINLEM). Experiments were carried out with the three proposed cluster based algorithms (CNLM, CNLEM and CINLEM) as well as the three existing algorithms (NLM, NLEM and INLEM). The results obtained with a set of medical images show that at low noise densities, the cluster based approaches outperforms the respective classical non local approaches. As the proposed approach can be easily parallelized, the extra time required for cluster formation can be justified.