Image Copy-Move Forgery Detection and Localization Scheme: How to Avoid Missed Detection and False Alarm

TL;DR

The paper addresses copy-move forgery detection and localization by tackling missed detections in low-resolution patches and false alarms from SGOs. It introduces an Excessive Keypoint CMFDL framework with a group-matching stage and a robust iterative localization that distinguishes tampering from self-similarity, achieving improved pixel-level localization. Key contributions include a $16\times16$-patch based excessive keypoint strategy with scale factor $s$, a three-step grouping-based matching, and a density-driven iterative localization with IQR-based grayscale verification. Experiments across six datasets demonstrate robust performance against rotation, scaling, and post-processing, with code released for public use.

Abstract

Image copy-move is an operation that replaces one part of the image with another part of the same image, which can be used for illegal purposes due to the potential semantic changes. Recent studies have shown that keypoint-based algorithms achieved excellent and robust localization performance even when small or smooth tampered areas were involved. However, when the input image is low-resolution, most existing keypoint-based algorithms are difficult to generate sufficient keypoints, resulting in more missed detections. In addition, existing algorithms are usually unable to distinguish between Similar but Genuine Objects (SGO) images and tampered images, resulting in more false alarms. This is mainly due to the lack of further verification of local homography matrix in forgery localization stage. To tackle these problems, this paper firstly proposes an excessive keypoint extraction strategy to overcome missed detection. Subsequently, a group matching algorithm is used to speed up the matching of excessive keypoints. Finally, a new iterative forgery localization algorithm is introduced to quickly form pixel-level localization results while ensuring a lower false alarm. Extensive experimental results show that our scheme has superior performance than state-of-the-art algorithms in overcoming missed detection and false alarm. Our code is available at https://github.com/LUZW1998/CMFDL.

Figures (8)

• Figure 1: Keypoint detection using (a) classical contrast threshold; (b) reducing contrast threshold; (c) upsampling and classical contrast threshold; (d) upsampling and reducing contrast threshold.
• Figure 2: Framework of the proposed scheme.
• Figure 3: Illustration of the grayscale-entropy clustering.
• Figure 4: Illustration of the new iterative forgery localization algorithm. (a) unvisited matches; (b) sampling set; (c) inliers with the same homography transformation; (d) suspicious region and (e) localization result in the iteration.
• Figure 5: Illustration of a sampling set with multiple models.