Robust image hashing via visual attention model and ring partition

Robustness is an important property of image hashing. Most of the existing hashing algorithms do not reach good robustness against large-angle rotation. Aiming at this problem, we jointly exploit visual attention model and ring partition to design a novel image hashing, which can make good rotation robustness. In the proposed image hashing, a visual attention model called PFT (Phase spectrum of Fourier Transform) model is used to detect saliency map of preprocessed image. The LL sub-band of saliency map is then divided into concentric circles invariant to rotation by ring partition, and the means and variances of DWT coefficients on concentric circles are taken as image features. Next, these features are encrypted by a chaotic map and the Euclidean distances between normalized encrypted features are finally exploited to generate hash. Similarity between hashes is measured by L 1 norm. Many experimental tests show that our image hashing is robust to digital operations including rotation and reaches good discrimination. Comparisons demonstrate that classification performance of our image hashing outperforms those of some well-known hashing algorithms in terms of receiver operating characteristics curves. Simulation of image copy detection is carried out on an open image database called UCID and the result validates effectiveness of our hashing.