Adaptive selection of search region for NLM based image denoising

Abstract

The non-local means (NLM) algorithm exploits the self-similarities or repeated patterns present in the whole image or a predefined search window for denoising the image. The size of the search window plays a crucial role in the performance of the NLM algorithm. If the search window used in the algorithm is larger than the required size, then it leads to over smoothing of the image whereas the choice of a smaller search window may result in inadequate noise removal. Therefore, ideally, the search window size must optimally vary from region to region based on the characteristics of the search region. The proposed algorithm selects an optimal size of search window for each pixel such that the variance of search region in the filtered image is close to the estimated variance of the corresponding region in an original image. The experimental results have shown that the proposed algorithm performs better than the original NLM and other state-of-the-art algorithms in terms of PSNR(dB), SSIM and visual quality for denoising the standard test images.