An Image Reconstruction Algorithm Using Patch-based Locally Optimal Wiener Filtering

Generally, the patch-based image processing methods use Euclidean Distance as the criterion of similar patches, which is impossible to fully reveal the patch structures and brings about the existence of blocking effect in the reconstructed image. In this paper, combining with the patch redundancy exploitation of the patch-based Wiener filter and the spatial distribution of blocking effect, an improved image reconstruction algorithm based on the patch-based locally optimal Wiener filtering is proposed. Firstly, the high-frequency part of the image is sampled sparsely, restricting the blocking effect to the border of adjacent blocks. Then the image is separately divided into two parts: the marginal region of blocks and the central region of blocks, and photometrically and geometrically similar patches are further utilized to determine the filtering parameters, which contribute to the smoothness of blocking region. As is shown by the experimental results, the proposed algorithm is able to efficiently reduce the blocking effect produced in the reconstruction and achieve a much better performance in terms of the images with rich textures.