An improved anisotropic diffusion filter with semi-adaptive threshold for edge preservation

• A new anisotropic diffusion filter based on a semi-adaptive threshold is proposed.
• A novel method with local difference value is applied to extract corrupted pixels.
• The improved method performs well in both edge preservation and noise removing.

This paper presents a noise removal method based on a semi-adaptive threshold in anisotropic diffusion filter to get better detail information protection and stronger noise suppressing ability. In this model, a method of local difference value is applied to distinguish corrupted pixels and noise-free pixels, and parts of the corrupted pixels are replaced by the pixels which have been pre-denoised through a Gaussian filter. Then an anisotropic diffusion model with a semi-adaptive threshold in diffusion coefficient function is applied to get a restored image. In order to implement the semi-adaptive threshold for each diffusion, the gradient value of the corrupted pixels is introduced in the threshold, which results in more diffusion in smooth areas and less diffusion in boundary regions. Compared with the traditional anisotropic diffusion models, the experimental results show that the proposed method can improve the PSNR by 30% and SSIM by 5%. The filtered images and experimental data indicate that the proposed method performs efficiently in both edge preservation and noise removing.