The Effect of Attacks on DCT-Based Features for Image Copy Detection

Nowadays images have been easily reproduced and distributed globally very fast. For the copyright protection, the copy detection is an important issue. Any modification on image copying processes can be viewed as attacks. The attacks include adding some small information as well as modifying images. In addition, the content-based copy detection techniques are relied on the image properties which could be extracted as features to be used in comparison between the original image and its copy. In this paper, we propose the DCT-based feature extraction technique which uses pixel values, edge, texture and frequency distribution of the image as features. A set of experiments have been conducted to evaluate several attacks in terms of robustness and sensitivity. The results show that when attacked, the image features react differently according to the types of attack. Different types of DCT-based features keep different piece of information and their robustness and sensitivity can show how features and attacks relate.