Improvement of BM3D Algorithm Based on Wavelet and Directed Diffusion

Block-matching and 3D filtering (BM3D) has shown great power for image denoising. By decomposing the BM3D restored image into three orthogonal subbands, we find that the subband of the high-frequency component, of which the wavelet coefficients are less than a given threshold, has made some true textures of the original image lost and, the PSNR of the restored image is pulled down. Therefore, in this paper, we propose a modified BM3D algorithm to solve these problems. We diffuse the subband of the high-frequency component, of which the wavelet coefficients are less than the threshold of the BM3D restored image to the corresponding subband of the noisy image to get a new estimate by using an improved directed diffusion equation model. The Laplace operator is replaced by an anisotropic diffusion operator and two different coefficients are added in the two diffusion terms. By replacing the corresponding subband of the BM3D restored image with the new estimate, we obtain the improved denoising result. Experimental results demonstrate that the proposed method achieves better performance than original BM3D algorithm in terms of both removing noise and persevering the edges and details of the image.