Abstract: Aiming at the problem that the image contaminated by high intensity salt and pepper noise, the author proposed a nonlocal switchable denoising algorithm. First, the noisy image was cut into several local blocks as the basic processing unit of the algorithm, and the prerepairing process of the blocks was designed to avoid the interference of the local blocks caused by the high noise pollution and ensure the similarity measure and the accuracy of the model estimation process. Then, a nonlocal filter was constructed in the nonlocal range, the author measured the similarity and calculated the repair weights to restore the noise, and obtained the omnidirectional image denoising effect. The experimental results show that compared with the traditional method of constructing filter based on local information, the proposed algorithm improves the performance and image measurement index of both lowpollution and highpollution images.

Key words: image denoising, salt and pepper noise, nonlocal filtering, Euclidean distance