Localizing JPEG Image Forgeries via SIFT and DCT Coefficient Analysis

As JPEG is one of the most popularly used image formats, localizing JPEG image forgeries plays a special role in passive forensics. In this paper, a copy-move forgery detection scheme combined scale invariant feature transform (SIFT) with discrete cosine transform (DCT) coefficient analysis in JPEG images is proposed. Generally, all the matching sift keypoints are located at the suspected regions obviously by means of sift-based forensic methods. But they cannot discriminate between original and forged regions. In order to localize the regional tampering, two conditional probabilities formulas made by histogram and interval length of quantized DCT coefficients are introduced. By comparing the two probabilities of 8 × 8 image block to which each sift keypoint belongs, it can determine if the block is authentic or not. Experimental results demonstrate that the proposed scheme is effective on localizing the tampered area, especially for avoiding misjudgment in a true image.