An enhanced switching median filter with neighborhood based layer discriminative noise detection for highly corrupted images

A novel enhanced switching median filter incorporating an effective impulse noise detection method called neighborhood based layer discriminative noise detection (NBLDND) is proposed in this paper for de-noising extremely corrupted images. The neighborhood of a pixel is classified into pepper corrupted lower class (PCLC), uncorrupted middle class(UCMC) and salt corrupted higher class (SCHC). The center pixel processed is considered as uncorrupted if it belongs to the middle class or otherwise it is labeled as corrupted and undergoes a filtering phase. The two layers that discriminate these three groups are accurately determined using NBLDND algorithm for yielding very high noise detection accuracy. The concept of adaptive window size and number of neighboring pixels accounted for filtering the corrupted ones which is suggested in this paper further makes the filtering more accurate. Extensive range of simulation results conducted on gray scale images under a wide range of (10% to 95%) noise corruption clearly shows that our proposed NBLDND algorithm (Neighborhood based Layer Discriminative Noise Detection) outperforms all existing median based filters in terms of filtering noise while preserving the image details. The results are compared with Boundary Discriminative Noise Detection algorithm (BDND) and finds that it comparatively works better for highly corrupted images.