Hierarchical Subspace Regression for Compressed Face Image Restoration

In this paper, we propose a novel hierarchical subspace regression algorithm based on edge orientations of compressed face image patches which includes two parts: training and restoration phases. In the training phase, the rule of the face edge-orientation (EO) distribution is used to classify the image patches into shallow subspaces. Then, the k-means clustering is used to cluster the deep subspaces of each EO-based shallow subspace, and corresponding linear mapping training is performed for each deep subspace. In restoration phase, an appropriate linear mapping selected based on the EO of compressed input image patch is applied to generate the restored output image patch. The experimental results show that the peak signal to noise ratio (PSNR) and structural similarity index (SSIM) are better than the existing popular algorithm, and they can effectively remove the blocking artifact and zigzag effect, so as to improve the visual effect.