Segmented-Based Region Duplication Forgery Detection Using MOD Keypoints and Texture Descriptor

Nowadays, with a rapid development of digital image technology, image forgery is made easy. Image forgery has considerable consequences, e.g., medical images, miscarriage of justice, political, etc. For instance, in digital newspapers, forged images will mislead public opinion and falsify the truth. In this paper, we proposed a segmentation-based region duplication forgery detection method, by extracting Maximization of Distinctiveness (MOD) keypoints for matching from segmented regions in the image. The main challenge is when the duplicated regions have been affected by rotation and scaling attacks. As a result, the proposed method detects duplicated regions based on two stages, structure analysis and texture analysis. In the first stage, the doubtful image is segmented into regions using the K-means algorithm. The segmented regions then labeled by centroids and MOD keypoints to represent their internal structures. MOD detects local interest points that are robust to rotation and improve detection rate in term of True Positive Rate (TPR). In the second stage, in order to identity the validated forged region, we explore Multiobjective Gradient Operator (MO-GP) to study the internal texture of segmented regions and eliminate the False Positive Rate (FPR) of forged regions. Experiment results show that our method can detect region duplication forgery under rotation, blurring and noise addition for JPEG images on MICC-F220 dataset with average TPR = 93% and FPR = 2%.