A switching median filter with boundary discriminative noise detection for extremely corrupted images

摘要: A novel switching median filter incorporating with a powerful impulse noise detection method, called the boundary discriminative (BDND), is proposed in this paper for effectively denoising extremely corrupted images. To determine whether current pixel corrupted, BDND algorithm first classifies pixels of localized window, centering on pixel, into three groups-lower intensity noise, uncorrupted pixels, and higher noise. The center will then be considered as "uncorrupted," provided that it belongs to "uncorrupted" group, or "corrupted." For that, two boundaries discriminate these groups require accurately determined yielding very high accuracy-in our case, achieving zero miss-detection rate while maintaining fairly low false-alarm rate, even up 70% corruption. Four models are performance evaluation. Extensive simulation results conducted both monochrome color images under wide range (from 10% 90%) corruption clearly show substantially outperforms all existing median-based filters, terms suppressing preserving image details, yet, algorithmically simple, suitable real-time implementation application.