A Modified Adaptive Hysteresis Smoothing Approach for Image Denoising Based on Spatial Domain Redundancy

Abstract

Hysteresis Smoothing (HS) is one of the recent denoising methods developed for images corrupted with additive white Gaussian noise, in which the Hysteresis process is used for noise reduction based on the image local characteristics. Many HS-based methods were introduced to improve the image denoising quality, while the local nature of this group of methods is always a limitation for their denoising performance. In this paper, considering the image data redundancy property, which expresses that each small block in an image could have many similar blocks in the same image, it is shown that by taking into account this property in the HS procedure, the quality of image denoising can be improved. According to that, a non-local adaptive Hysteresis Smoothing (NLAHS) approach is introduced in such a way that the denoised value of the under-processing pixel can be estimated by weighing and averaging the values obtained from the HS process in the non-local blocks, which are more similar to the under-processing pixel reference block. Also, to achieve a more reliable method for similarity determination between the blocks, a path-based NLAHS (PBNLAHS) method is introduced. In this method, the weighting process is modified by considering scanning paths instead of peer-to-peer differences for block similarity determination. The comparative simulation results of the proposed methods show their significant improvement over the previous HS-based methods and some other well-known denoising approaches in terms of both objective and subjective criteria.