Improved Median Filtering for Image Denoising

Abstract

Traditional median filters easily blur fine textures and weaken edge definition during salt-and-pepper noise elimination. To tackle this limitation, this paper proposes an improved median filter integrated with adaptive thresholds and local statistical features. The algorithm firstly classifies the central pixel as an extreme point or ordinary pixel. For extreme pixels, median values and adaptive thresholds are computed only from valid non-extreme pixels inside the filtering window; for non-extreme pixels, dynamic discrimination thresholds are built relying on the global window's median and local deviation. A central pixel will be substituted with the corresponding median once its gray deviation exceeds the adaptive threshold. Simulation results demonstrate that the presented method achieves remarkable superiority over conventional median filters across multiple noise densities, verifying its reliability for image denoising tasks.