Image Copy Detection with Rotation and Scaling Tolerance

Currently, most of image copy detection methods can successfully resist to the noise-like distortion, but they are quite fragile to geometric distortion, such as rotation, shift, translate, scale, cropping and so on. Among the geometric distortions, rotation and scaling most commonly happen. In order to really resist against rotation, shift, scale and crop distortion, Wu et al. proposed an ellipse track division based image copy detection method in 2005 and 2007 respectively. However, because of ellipse shape without rotation invariance property, Wu's method didn't really address rotation distortion problem. To completely conquer the rotation and scale distortion issues, the authors propose a novel image copy detection scheme which combines cirque division strategy with ordinal measure method to extract compact image feature. It is well known that the content within cirque track region will be almost invariant before and after rotation and scale distortion, so the proposed method can successfully resist the above-mentioned two kinds of distortion. In addition, since the ordinal measure based feature vector is insensitive to local or global slight changes among image content, so the noise-like distortion can be effectively conquered. The experimental results show the proposed method is better than that of Wu at the aspect of rotation and scale distortion.