Wave Atom-Based Perceptual Image Hashing Against Content-Preserving and Content-Altering Attacks

This paper presents a perceptual image hashing algorithm based on wave atom transform, which can distinguish maliciously attacked images from content-preserving ones. Wave atoms are employed due to their significantly sparser expansion and better feature extraction capability than traditional transforms, like discrete cosine transform (DCT) and discrete wavelet transform (DWT). Thus, it is expected to show better performance in image hashing. Moreover, a preprocessing method based on Fourier-Mellin transform is employed to keep the proposed scheme against geometric attacks. In addition, a randomized pixel modulation based on RC4 is performed to ensure the security. According to the experimental results, the proposed scheme is sensitive to content-altering attacks with the resiliency of content-preserving operations, including image compression, noising, filtering, and rotation. Moreover, compared with some other image hashing algorithms, the proposed approach also achieves better performance even in the aspect of robustness, which is more important in some image hashing application, for example image database retrieval or digital watermarking.