Image authentication based on robust image hashing with geometric correction

Image authentication based on robust image hashing has been widely discussed with continuous improvements. However, most of the existing methods misjudge the images processed through the geometric transformation and small malicious operations. In this paper, we have proposed a geometric correction approach, which eliminates the influence of geometric transformation, including composite rotation-scaling-translation (RST). We have incorporated the local features with global features to construct hash. The local features are extracted from the salient regions, which have been obtained using Markov absorption probabilities. The global features include statistical feature distance. While being robust to content-preserving operations (including the composite RST) the hash is sensitive to the small malicious operations, and hence may be used for image authentication. The proposed image authentication is a two-phase system. First, the threshold, δ, based on only global feature segregates between “different (also includes large area tampering) images” and “similar images or small tampered images”. In the next phase, the similar image pairs are authenticated and the tampered regions are localized, based on local features. The receiver operating characteristics shows that the proposed image authentication system outperforms some state-of-the-art methods.