Normalized Moment of Inertia-Based Detection Algorithm for Copy-Paste Image Tampering

Abstract

Copy-paste tampering is one of the most common image content attacking methods. Considering the low accuracy and high feature dimension of the existing algorithms, a normalized moment of inertia method is proposed in this paper to overcome these defects. In the phase of feature extracting, the algorithm first transforms the tested image using wavelet, and then selects the similar subbands to build overlapped blocks, finally uses Perceived Hash Algorithm (PHA) to make binarization processing for the subblocks and carries out the normalize moment of inertia of the subblocks which satisfy the coarse matching conditions between adjacent two lines after performing dictionary sorting. In feature matching phase, the algorithm first counts the similar subblocks whose shift is above the distance threshold, and then obtains the main shift vectors with specific frequencies, finally performs feature matching according to the difference of the normalized moment of inertia in the neighborhood. Experiment results illustrate that the proposed algorithm with lower feature dimension can efficiently improve the matching speed and accuracy. Furthermore, it has better robustness for some post-processing operations, such as compression, Gaussian Blur, adding Gaussian noise, etc.