Image forgery detection based on the fusion of machine learning and block-matching methods

摘要: Dense local descriptors and machine learning have been used with success in several applications, like classification of textures, steganalysis, forgery detection. We develop a new image detector building upon some recently proposed the steganalysis field suitably merging such descriptors, optimizing SVM classifier on available training set. Despite very good performance, small forgeries are hardly ever detected because they contribute little to descriptors. Therefore we also simple, but extremely specific, copy-move based region matching fuse decisions so as reduce missing detection rate. Overall results appear be encouraging.