Laplacian-preprocessed impulse-noise detection, with image denoising via difference-mean-filtering of long-range-correlated sub-images

Zhang & Karim's Laplacian-preprocessed detector (2002) is robust against mis-identification of an image's thin-lines as impulse-noise-corrupted pixels. Wang & Zhang's "long-range correlation" denoising scheme (January 1998) exploits any information-redundancy between an identified corrupted-pixel's local neighborhood with distant sub-images, to restore the corrupted pixel. This paper synergizes the above two algorithms, with the following algorithmic enhancements: (1) a pre-tuning of Zhang & Karim's threshold based on a rough estimation of the corrupting impulse-noise's spatial probability of occurrence, assuming the availability of a test-image "sufficiently" similar to the given corrupted image; and (2) a new "difference-mean" criterion for better pixel-restoration. Limited simulations illustrate the above proposed scheme's efficacy and improvements.