Detecting Copy-Move Forgery Using Non-negative Matrix Factorization

Abstract

Copy-move is a common manipulation of digital image tampering. Block matching after feature extraction is mostly applied method to detect this kind of forgery. In this paper, we propose an efficient copy-move detecting scheme with the capacity of some post-processing resistances. The image is divided into fixed-size overlapped blocks, and then non-negative matrix factorization (NMF) coefficients are extracted from list of all blocks. It is worth noting that all the coefficients are quantized before matching, in other words, we could use relatively few data to interpret a sub-image. Besides, we apply lexicographical sorting method to reduce the probability of invalid matching. Lastly we measure the hamming distance of each block pair in the matching procedure, if the distance is shorter than a threshold, we localize them as the tampering areas. Experimental results show the efficacy of our scheme.