Efficient Restoration of Magnetic Resonance Images Corrupted with Impulse Noise using Spatial Constraints based Fuzzy Decision Filter

SPATIAL CONSTRAINT BASED FUZZY DECISION FILTER

Abstract

Magnetic Resonance (MR) images are subject to unavoidable noises during the data acquisition due to imperfections of device components and trade-offs in the scan parameters. The study proposes a two-step Fuzzy Decision-Based Filter (FDBF) as a post-reconstruction technique to mitigate random valued impulse noise from MR images. The FDBF employs a Spatial Fuzzy C-means (SFCM) clustering for detection and an Intensity Based Fuzzy Estimation (IBFE) technique for restoration. Firstly, SFCM integrates the spatial relation of the adjacent pixels into the membership function to form three separate clusters. The IBFE technique leaves the noise-free cluster unaltered while restoring the remaining in the second step. IBFE incorporates neighbor pixel correlation to restore the corrupted pixel leading to edge preservation. To assess the efficacy of the intended method both the quality metrics and the observed quality of the restored images are considered. The suggested detection strategy using SFCM performs very well, up to a 93% corruption level with zero false and miss detection rates even when there is intensity in homogeneity among pixels. Compared to other existing filtering techniques, the proposed two-step restoration method significantly improves the perceived image quality and other image quality metrics of the restored image without obliterating more intricate details and finer structures. FDBF considers the spatial information of the nearby pixels during the detection and restoration processes, which is essential for MR image restoration.