Image hashing algorithm based on randomised log-Gabor filtering and sparse random projection for image copy detection

This study proposes a lightweight and secure image hashing algorithm combining randomized log-Gabor filtering with sparse random projection (SRP) for image copy detection. Unlike methods relying on fixed filter banks or deep learning, our approach extracts texture features using log-Gabor filters whose orientation parameters are randomized via a logistic chaotic map for enhanced security. Variance features capture local structural details, and SRP compresses the feature matrix while preserving distances per the Johnson-Lindenstrauss lemma. Experiments on public datasets demonstrate strong robustness against common attacks. The algorithm achieves approximately 123 fps on a standard CPU with over 99% less memory than deep-learning methods, suiting resource-limited cyber-physical systems.