An Image Fingerprinting Method Robust to Complicated Image Modifications

Xinghua Yu, Tiejun Huang

Abstract: Image fingerprinting technique plays an important role in digital rights management and related applications, especially in the detection of the illegal use of image works. Although existing image fingerprinting methods perform well in many cases, they are not robust enough to sophisticated modifications like image embedding and image combining. This paper proposes an image fingerprinting approach which is robust to such modifications. Firstly, we introduce an SIFT-based algorithm to extract image keypoints as the unique fingerprint. Secondly, a technique based on geometry isomorphic relationship is utilized to select valid keypoints in the queried image. Finally, by calculating the matched keypoints pairs between the queried image and the pre-registered image, we can make the decision whether they are homologous. Experimental results show that the proposed method is robust to image embedding and combining, without a high computational complexity.