Perceptual image hash function via associative memory‐based self‐correcting

Abstract

Paralleling with the revolutionary development of the Internet, there has been increasing concern about the copyright infringement of digital media. A central problem in copyright protection is to accurately and efficiently identify the illegal copies of copyrighted contents. Perceptual hash function, which summarises the perceptual characteristics of digital media to a short digest, is a low-cost solution to this problem. Owing to the easy-to-manipulate nature of digital media, a major challenge in designing perceptual hash function is to achieve the robustness against distortion. To tackle this problem, an associative memory-based hash function is introduced. The proposed work repairs the distortions on local image structures via associative memory-based de-noising, in the hope of simulating the self-correcting mechanism in human memory. The shape invariant descriptors of de-noised structures are then encoded to binary bits. Experimental results confirm that the proposed work outperforms representative algorithms, and it can achieve an equal error rate of 3.30 × 10 − 3 in content identification with only 80 bits.