Adaptive Edge-guided Block-matching and 3D filtering (BM3D) Image Denoising Algorithm

Image denoising is a well studied field, yet reducing noise from images is still a valid challenge. Recently proposed Block-matching and 3D filtering (BM3D) is the current state of the art algorithm for denoising images corrupted by Additive White Gaussian noise (AWGN). Though BM3D outperforms all existing methods for AWGN denoising, still its performance decreases as the noise level increases in images, since it is harder to find proper match for reference blocks in the presence of highly corrupted pixel values. It also blurs sharp edges and textures. To overcome these problems we proposed an edge guided BM3D with selective pixel restoration. For higher noise levels it is possible to detect noisy pixels form its neighborhoods gray level statistics. We exploited this property to reduce noise as much as possible by applying a pre-filter. We also introduced an edge guided pixel restoration process in the hard-thresholding step of BM3D to restore the sharpness of edges and textures. Experimental results confirm that our proposed method is competitive and outperforms the state of the art BM3D in all considered subjective and objective quality measurements, particularly in preserving edges, textures and image contrast