A keypoint based copy-move forgery detection and localization in digital images / Somayeh Sadeghi

• Main Author: Somayeh, Sadeghi
• Format: Thesis
• Published: 2015
• Subjects: QA76 Computer software; T Technology (General)
• Online Access: http://studentsrepo.um.edu.my/6141/1/Final_Thesis.pdf; http://studentsrepo.um.edu.my/6141/

Due to the fast development of powerful image processing tools and the importance of image integrity, digital image forgery has become a very important topic for certain organizations. Copy-move forgery is one of the most commonly used types of digital image forgery, where one part of the image is copied and placed elsewhere in the same image. Because of the existence of various digital environments, a copy-move forgery detector should be robust against pre- and post-processing operations, such as scaling, rotation, JPEG compression and noise. A copy-move forgery detector should be able to detect forgery in a reasonable amount of time. In this research, an image authentication scheme with the capability of copy-move forgery localization is proposed, based on the scale invariant feature transform (SIFT). The importance of the proposed method is its ability to authenticate digital images and accurately locate copied and pasted areas. The proposed algorithm starts by extracting local image features, which are known as keypoints, using SIFT, followed by searching for similar keypoints by clustering extracted descriptors from the image. Finally, matched keypoints, which are duplicated regions in the image, are connected to each other to illustrate which part of the image has been tampered with. Several experiments are performed to validate the effectiveness and robustness of the proposed algorithm against different attacks, such as pre-processing attacks. The experimental results illustrate that the proposed algorithm is robust against several geometric changes, such as JPEG compression, rotation, noise and scaling. Furthermore, the detection rate of the algorithm is improved by utilizing the proposed clustering procedure. The true and false positive rates achieved by the proposed algorithm outperform several current detection algorithms.