Block Mean Value Based Image Perceptual Hashing for Content Identification

Image perceptual hashing has been proposed to identify or authenticate image contents in a robust way against distortions caused by compression, noise, common signal processing and geometrical modifications, while still holding a good discriminability to tell the same content from different ones in sense of human perception. We propose and compare in this paper four normalized block mean value based image perceptual hashing algorithms which demonstrate higher performances than other existing algorithms in robustness-and-discriminalibility and simplicity for implementation. Overlapped blocking and complexity-reduced rotation operations are employed to enhance the robustness to geometrical distortions, especially to small to middle degree rotations. Given fixed original image contents, identification ratio is used to evaluate the algorithms’ content identification. Given fixed content classification, receiver operating curves is obtained to evaluate the proposed algorithms’ content robustness and discriminability based on inter-and intra-content hashing distances.