A new learning-based deblocking algorithm for DCT coded images

A new algorithm for the reduction of blocking artifacts in images compressed using block-based discrete cosine transform (DCT) is proposed in this paper. Firstly, a Bayesian model and the Markov network assumption are adopted for our deblocking algorithm. An input blocking image is divided into observation nodes of the network. Then a simplified method is applied to find approximate optimal solutions of the underlying nodes in the network. The solutions are learned from the training set. Experimental results show that the proposed approach is able to remove some blocking artifacts, at the same time, reserve sharp edges and learn fine details.