The invention belongs to the image processing field and relates to an adaptive anisotropic filtering method. With the method adopted, a gradient threshold value can be changed according to local statistical features, and in anisotropic diffusion, the weights of different pixels are dynamically adjusted, so that the denoising ability of anisotropic filtering in a high-intensity noise environment can be improved. According to the local statistical feature-based adaptive gradient threshold anisotropic filtering method provided by the technical schemes of the invention, based on an anisotropic filtering algorithm, a gradient operator is utilized to identify image gradient change caused by noises and image gradient change caused by edges, neighborhood weighted averaging is adopted to remove small gradient change caused by noises and retain large gradient change caused by edges, and the above process is carried out iteratively until noises in an image are removed. The method of the invention is mainly applied to image processing.