A Novel Perceptual Image Hashing Method via Geometric Features and Distance Invariant

A novel image perceptual hashing method is proposed on the intermediate hashing stage. It first uses the iterative geometric techniques to get main geometric features. Then, by the observation that the distances of feature points are invariant in the polar coordinate under arbitrary rotation, a novel processing method is proposed to arrange the two-dimensional distribution to a one-dimensional feature vector. It is verified by our detailed experiments that the proposed method can withstand standard benchmark (e.g. Stirmark) attacks. Moreover, the projection processing makes the iterative geometric techniques have a stronger robustness under rotation attack, which is more than 15 degree for most images and can be applied to any low-level image feature extraction approaches as well to improve the rotation robustness. At last, the factors which influence the performance are analyzed and the further steps to improve the rotation robustness are also given.