Copy-Move Forgery Detection Utilizing SIFT and FLANN with Optimized Keypoint Matching

The safety and authenticity of digital images have become more difficult to guarantee with the development of image processing tools which makes image forgery a serious issue. In this paper, a method for detecting copy-move forgery in digital images is proposed. Scale Invariant Feature Transform (SIFT) is used to extract keypoints and corresponding descriptors. Fast Library for Approximate Nearest Neighbors (FLANN) is employed to match keypoints where both spatial and feature distances are considered. Additionally, the Lowe ratio test is applied to exclude ambiguous pairs of matched keypoints. Forgery is identified if the number of pair of matched keypoint is greater than or equal to a predefined threshold value. Fixed-size small patches are considered around each detected keypoints and the corresponding pixels of the patches are colored white or green in a black image. This approach highlights the source and forged regions in different colors. The MICC-F220 and MICC-F2000 datasets are used to conduct experiments to determine the threshold value for the minimum number of matched keypoint pairs. The optimized value for this threshold is found to be 2. According to the proposed method, if at least 2 pairs of keypoints are matched in an image, the image is classified as forged. Evaluation metrics including Recall, Precision, False Positive Rate (FPR), Accuracy, and F1-Score are calculated. F1-Score of 95% and 81% are achieved for the MICC-F220 and MICC-F2000 datasets respectively.