Non-Local Means Image Denoising Algorithm Based On Edge Detection

As the basis of image processing, image denoising has always been a hotspot in the field of computer vision. The purpose of image denoising is to restrain the noise from the anamorphic image and restore it as much as possible, making the restored image closer to the original image qualitatively. Although there are numerous kinds of noises, this thesis is mainly aimed at the removal of gaussian noise and salt-and-pepper noise.Aiming at the removal of gaussian noise, we systematically analyze the shortage of non-local means image denonising algorithm (NLM), finding it is easy to lose structure information when dealing with the image containing complex edges and textures by NLM algorithm. In order to solve this problem, a non-local means image denoising based on edge detection is proposed in this thesis. The innovation of the proposed algorithm is mainly manifested in the following:(1) An improved Sobel operator with eight directions is proposed to extract a more accurate edge image; (2) To make the neighborhoods with similar structure obtain more weight, not only the Euclidean distance but also the edge image are considered when the similarity of neighborhoods is measured. Many experiments demonstrate that in both subjective and objective evaluation principles the performance of the improved algorithm has a good effect, and the visual effect of the denoised image is good.Aiming at the removal of gaussian noise, a noise adaptive switching non-local means denoising algorithm (NASNLM) is proposed in this thesis. For noise detection, the pixels of image are divided into the noise and the non-noise points. For filtering, four different filtering techniques are adopted:switching filtering, noise adaptive median filtering, edge-perserving filtering and non-local means filtering. Switching filtering can keep the gray-value of non-noise points unchanged. Noise adaptive median filtering can suppress the high-density salt-and-pepper noise. Edge-preserving filtering can preserve more image edges and details. Non-local means filtering can further improve the ability of noise suppression and detail maintenance. Experiments demonstrate that for removal of the high-density salt-and-pepper noise by NASNLM algorithm, a better denoising effect is obtained than other methods.