An improved method to image copy detection

Kim proposed a novel method to detect image copies based on DCT coefficients and ordinal measure. This method had a good ability to resist various signal and local geometric manipulations. However, Kim´s method was based on block units, which would lead to an insufficient precision. In order to solve the problem, on the basis of Kim´s method, we use an extended feature set which takes advantage of prior information to detect more attack types and the comparison of fine-grained edge information which can improve retrieval precision to further perfect Kim´s method. Experimental results show the improved method not only detects all copies that Kim can detect, Such as histogram equalization, contrast enhancement, water coloring, motion blurring, mosaic tiling, Gaussian noising, resizing, rotate with 180⁰ and so on, but also significantly improves retrieval precision.