JPEG image tampering localization based on normalized gray level co-occurrence matrix

To locate the tampered region of double compressed JPEG images, one of the most effective methods is based on the statistical characteristic of the images. After tampering operation, the tampered region and the original region will have different statistical distributions. And according to this cue, the histogram of DCT coefficients can be moded as the mixture of the distributions of DCT coefficients in tampered and untampered regions. By estimating each distribution of the mixture model, the probability of being tampered of each DCT block can be calculated and the final localization result can be obtained. Thus the mixture model will significant impact the result. In this paper, a novel mixture model based on normalized gray level co-occurrence matrix (NGLCM) is proposed for tampering localization in JPEG images. Firstly, NGLCM is used to measure the conditional probabilities of being tampered regions and being untampered regions for each 8 × 8 DCT coefficient block, which can take advantage of both the statistical characteristic of double quantization effect and the relationship among neighboring blocks. Then the Bayesian posterior probability map is generated by the conditional probabilities, which indicates the tampering probability of each block. Finally, the map is refined based on a inter-block connectivity and Gaussian weighted filter strategy to determine the final tampered region location. Experimental results demonstrate that the proposed approach can localize the tampered region of JPEG images with a satisfactory performance and outperforms the state-of-the-art methods.