3D Filtering of Images Corrupted by Mixed Additive-Impulsive Noise

Abstract: A procedure for filtering images corrupted by mixed (additive-impulsive) noise has been substantiated and implemented for the first time. The novel method is characterized by the following stages: the detection and filtering of pixels corrupted by noise impulses, the image filtering in three-dimensional (3D) discrete cosine transform (DCT) space, and the final image processing stage in which the errors of the previous stages are corrected and the image edges and details are reconstructed. A physical interpretation of the filtering procedure under mixed noise conditions is given, and a filtering block diagram has been developed. Simulations of the proposed image filtering method have confirmed the advantage of the novel filtering scheme in terms of generally recognized criteria: the structural similarity index measure and the peak signal-to-noise ratio as well as when visually comparing the filtered images.