关键词: 图像去噪, 非局部均值, 结构检测, 边缘

Abstract: Concerning the problem of residual structure on Non-Local Means (NL-Means) image denoising algorithm, an improved NL-Means image denoising algorithm with structure detection was proposed. The noisy image was first processed by using a structure analyzer to extract detail information before filtering, then the similarity measure was modified to incorporate the result of edge detection, more weight was given to a pixel if its edge content was more similar to that of the pixel being denoised and the neighborhood with a dissimilar edge pattern should receive a lower weight (or zero) to preserve the original edge content. The experimental results show that the proposed algorithm enhances NL-Means denoising ability and image structural similarity, and improves the visual quality of the denoised image.

Key words: image denoising, Non-Local Means (NL-Means), structure detection, edge