Vision-based Anti-forgery of Chinese Paintings using MRF

摘要

This paper proposes a hierarchical framework for computer-aided forgery authentication of traditional Chinese painting using Markov Random Field. It is to determine whether a painting is forged, denoting by giving the label as the authentic or spurious. Firstly, we manually choose some patches from featured objects in paintings. The patches could be typical objects frequently appeared in Chinese traditional paintings, such as rocks, tree trunks etc. Then, based on these different kinds of patches, we construct the framework at three levels, i.e. pixellevel,patch-level and object-level. At pixel level, we formulate the problem of label determination as inference in MRF using belief propagation algorithm. The framework hierarchically provides an inferring processing to authenticate whether the painting is a forgery or not. Experimental results demonstrate the significant performance of our method.