Block Cluster Based Dictionary Learning for Image De-noising and De-blurring

摘要

Image de-noising or de-blurring is an important step in image preprocessing. A great variety of experiments have demonstrated that using image block as a basic operation unit can effectively improve the final results both in efficiency and visual quality. An image block searching algorithm based on the largest variance of inter groups is proposed by referring to HVS. This method could effectively extract the intrinsic information of the image blocks and avoid the change of the Euclidean distance due to the illumination variations. With the better variance value among different image block groups, the correlation of those groups is reduced and a dictionary of wider distribution is obtained such that it can get a better visual effectiveness in the sparse reconstruction. The experimental results show that this method outperforms state-of-the-art algorithms both in visual quality and the PSNR value.