The classical Perona-Malik anisotropic diffusion model can not remove the image noise while edge preserving effectively because it just considers the gradient information of the image. A novel modified anisotropic diffusion denoising scheme was proposed based on the correlation coefficient, which incorporates both the local gradient and the local gray variance information of the pixels in an image. Experimental results show that the proposed model can obtain better noise removing and edge keeping simultaneously than the traditional Perona-Malik model, and it has some generality and a certain application value in the image denoising.