Combination Of Sift Feature And Convex Region-Based Global Context Feature For Image Copy Detection

The conventional content-based image copy detection methods concentrate on finding either global or local features to handle the copy detection task. Unfortunately, the global features are not robust to the cropping attack, while the local features cannot substantially capture context information and thus are not discriminative enough. To address these issues, this paper proposes a novel image copy detection method, which combines both the global and the local features. Firstly, SIFT (scale invariant feature transform) features are extracted and then initially matched between images. Secondly, the SIFT matches are verified by the proposed convex region-based global context (CRGC) features, which describe the global context information around the SIFT features, to effectively remove the false matches. Finally, the number of the surviving SIFT matches is used to determinate whether a test image from image databases is a copy of a given query image. Experimental results have demonstrated the effectiveness of our proposed method in terms of both robustness and discriminability.