A robust forgery detection algorithm for object removal by exemplar-based image inpainting

Object removal is a malicious image forgery technique, which is usually achieved by exemplar-based image inpainting in a visually plausible way. Most existing forgery detection approaches utilize similar block pairs between inpainted area and the rest areas, but they invalidate when those inpainted images are further subjected to some post-processing operations such as JPEG compression, Gaussian noise addition and blurring. It is desirable to develop a forensic method which is robust to object removal with post-processing. From some preliminary experiments, we observe that post-processing destroys the similarity of block pairs and simultaneously disturbs the correlations among adjacent pixels to some extent. Inspired by the strong ability of joint probability density matrix (JPDM) in characterizing such correlation, we propose a hybrid forensics strategy. Firstly, our earlier method is employed to detect whether a candidate image is forged or not. Secondly, for those undetected images after the first step, JPDM is computed for each difference array to model the correlations among adjacent DCT coefficients, and the average of these matrixes are computed as feature vectors to further expose tampering traces. Experimental results show that the proposed approach can effectively detect object removal by exemplar-based inpainting either with or without post-processing.