Abstract: In the anisotropic diffusion filtering method, the accuracy of edge detection has an important influence on the filtering results. In order to enhance the image filtering effect, this paper put forward a novel anisotropic filtering method of high-fidelity based on threshold optimization. Firstly, wavelet is used to extract image high frequency part, second-differential modulus is used to express local information, this will avoid mistaking peak and corner for noise, and image local information will be saved. Then, least mean square algorithm is used to design a threshold, it will control the diffusing strength, and the novel anisotropic model will be established. Last, the novel model is used to process the high-frequency, and wavelet is used to reconstruct denoised image with the processed high frequency part and original low frequency part. Experiment results indicate that the novel method has a perfect performance, it can protect image details well. Moreover, the running time of the proposed method is relatively short, it is applicable, so the proposed method is ideal.