A Robust Image Copy Detection Scheme Using Ordinal Measure of Full DCT Coefficients

To effectively detect copyrighted images suffering both single processing attacks and geometrical attacks is one of the hotspots and difficulties in the image copy detection research. For most existing block-based image copy detection algorithms, the performances can successfully resist to the noise-like distortion, but they are quite fragile to geometric attacks, such as rotation, shift, translate, scale, cropping and so on, because the content of each block is changed after these kinds of attacks. In this paper, a robust image copy detection algorithm based on ordinal measure of the full DCT coefficients is proposed. The ordinal measure of the selected low and middle frequency coefficients which are calculated from full DCT transformation on Y-component of YCbCr color space is used as the image eigenvector. The middle and low frequency full DCT coefficients are changing regularly with scaling, so the proposed method can resist scaling distortion. Meanwhile, the ordinal measure is introduced to resist single processing attacks. Moreover, the rotation compensation strategy used in this algorithm can resist the rotation attacks in which the angles are less than 30°. The experimental results show that the algorithm has higher precision and recall rate than the existing methods in terms of rotation and scaling distortion.