Dual-stage Enhanced Decision based Unsymmetric Trimmed Mean-Median Filter for High Density Salt and Pepper Noise Removal

Abstract

This paper presents a high-performance enhanced decision based unsymmetric trimmed mean-median filter (EDBUTMMF) to detect and remove high density salt and pepper noise from an image. Unlike conventional decisionbased filters that directly classify all extreme-valued pixels as noisy, the proposed method first verifies whether an extreme-valued pixel is genuinely corrupted before restoration. The process of denoising is implemented in two stages: noise detection followed by noise removal. The proposed filter initially detects noisy pixels iteratively through several phases and replaces the noisy pixels with the trimmed method. Unsymmetric trimmed filtering omits impulse noise values before performing the median calculation, ensuring that the replacement pixel is derived exclusively from the uncorrupted neighbourhood samples. This yields improved edge preservation and restoration accuracy, especially when dealing with high noise densities images. Dual-stage method separates noise detection and restoration to avoid unnecessary modification of uncorrupted pixels, especially at high density noise. Performance evaluation shows both quantitative parameters MSE, PSNR and SSIM achieves better results, while providing improved visual quality and lower computational time compared with several existing decisionbased filtering methods.