Non-local means denoising based on SVD basis images

With the assumption that natural images contain considerable amount of self-similarity, non-local means image de-noising uses patches similarity in order to filter noisy images. Although the output of the Non local means algorithm is very desirable in removing the low level of noise, when the noise increases, the performance deteriorates. This is because the similarity cannot be evaluated perfectly through noisy patches. To solve this problem, in the proposed approach, the similarity evaluation for each patch is performed based on the structural information. This is due to the fact that the HVS (Human Visual System) is highly adopted to extract structural information from a viewing scene. In this paper, a modified non-local means filter is introduced in order to find better similar patches especially in the case of high level of noise. The experimental results show appropriate performance of the presented algorithm both visually and quantitatively.