Perceptual hashing in the 3D-DWT domain

Abstract

Arobust or perceptual hash function computes hashes of videos based on their perceptual contents. Two basic requirements for such hash function are the robustness against content-preserving operations and the sensitivity to dissimilar contents of distinct videos. This paper presents a new hash function that applies the 3D discrete wavelet transform on a group-of-frames and extracts a hash of it from the spatio-temporal low-pass band. The discrete cosine transform is applied on the 1D representation of the spatio-temporal low-pass band for decorrelation, and the transform coefficients in the mid-frequency region are used for hash computation. Experimental results show the robustness of the hash function against the content-preserving operations and the sensitivity to content differences in distinct videos.