Robust perceptual color image hashing using randomized hypercomplex matrix factorizations

摘要: Compact representations of color image and video content allow efficient search, retrieval storage this over the Internet online repositories. However, most these neither take into account inherent correlation nor perceptual redundancy information. In paper, we propose a hash representation for images using robust features. These features, dominant singular vectors extracted quaternion value decomposition (QSVD) pseudorandomly selected overlapping blocks, are efficiently used search applications. Their robustness is guaranteed by underlying vectors. The motivation behind our work twofold: 1) ability QSVD algorithm to provide best low-rank approximation in Frobenius norm sense 2) compact handle components as single entity. leads proper modeling possible geometric attacks an independent identically-distributed (i.i.d) quaternionic random noise on Such simplifies code detector design. Hash against evaluated large set test where proposed scheme outperforms existing factorization-based hashing algorithms terms lower miss false alarm probabilities orders magnitude. Finally, improved performance does not come at expense increased computational complexity which another salient feature scheme.