High Density Noise Removal by Cascading Algorithms

TL;DR

This paper introduces a two-stage cascading filter algorithm that effectively removes high-density salt and pepper noise from digital images, improving quality across all noise levels.

Contribution

It combines decision-based median filtering with advanced noise removal filters to enhance image quality at various noise densities, outperforming existing methods.

Findings

Better MAE, PSNR, and IEF than existing filters

Effective noise removal at all noise densities

Preserves image quality and reduces blurring

Abstract

An advanced non-linear cascading filter algorithm for the removal of high density salt and pepper noise from the digital images is proposed. The proposed method consists of two stages. The first stage Decision base Median Filter (DMF) acts as the preliminary noise removal algorithm. The second stage is either Modified Decision Base Partial Trimmed Global Mean Filter (MDBPTGMF) or Modified Decision Based Unsymmetric Trimmed Median Filter (MDBUTMF) which is used to remove the remaining noise and enhance the image quality. The DMF algorithm performs well at low noise density but it fails to remove the noise at medium and high level. The MDBPTGMF and MDUTMF have excellent performance at low, medium and high noise density but these reduce the image quality and blur the image at high noise level. So the basic idea behind this paper is to combine the advantages of the filters used in both the stages to remove the Salt and Pepper noise and enhance the image quality at all the noise density level. The proposed method is tested against different gray scale images and it gives better Mean Absolute Error (MAE), Peak Signal to Noise Ratio (PSNR) and Image Enhancement Factor (IEF) than the Adaptive Median Filter (AMF), Decision Base Unsymmetric Trimmed Median Filter (DBUTMF), Modified Decision Base Unsymmetric Trimmed Median Filter (MDBUTMF) and Decision Base Partial Trimmed Global Mean Filter (DBPTGMF).