Residue-based fusion of denoised images by different filter banks

The paper presents a new method to improve image quality that exploits the residue-based fusion of three denoised images using different two-dimensional (2-D) filter banks. First, the three denoised images are obtained by applying the BLS-GSM algorithm to the decompositions of a noisy image by the steerable pyramid, nonsubsampled contourlet, and 2-D separable oversampled DFT filter bank. The denoised images exhibit inhomogeneous quality relevant to local details of the image and the underlying filter bank. Second, the local quality of denoised images is measured by the local variances of the residual image, the noisy image minus the denoised image. Third, the three denoised images are fused into a clean image with the help of the local quality measures. The experimental results show that the proposed method improves the quality of denoised images visually and in the sense of peak signal-noise-ratios (PSNR).