Image denosing by curvature strength diffusion

Due to its great advantage that it can preserve image edges while reducing noise, the anisotropic diffusion open a new filed in image processing. However, as anisotropic diffusion is based on gradient, which is sensitive to noise, it may not work efficiently especially when the image contains a high level of noise. In this paper, a new method is proposed to tackle this problem. Making use of the local analysis of an image, Hessian matrix, we propose a new idea of curvature strength to describe the intensity variations in images. Employing the curvature strength to tune the diffusion, the proposed diffusion scheme works better than the original anisotropic diffusion. Experimental results on several standard images demonstrate that the proposed scheme has a better and more robust denoising ability than the original anisotropic diffusion.