Copy-move forgery detection on digital images

One of the most common types of forgery on digital images is copy-move forgery. In the thesis study, the problems in copy-move forgery detection methods are examined and new methods are proposed to detect forged images to overcome these problems. In the first of the studies, the image was evaluated from the L*a*b* color space and the forged regions were determined roughly by the keypoints extracted from each color channel. A dynamic localization step is proposed, which provides independence from the input image, in delineating the precise boundaries of the forged region. In the second proposed method, a texture image that preserves the rotation-invariant feature of the image is obtained. Suspicious regions were created by matching the keypoints obtained from the whole texture image with each other. A Ciratefi- based approach is presented in determining the forgery limits in the roughly forged regions. With the approaches proposed in the thesis, robustness against different attack types (rotation, scaling, JPEG compression, noise addition) is provided. Thus, forged images were detected with the accuracy rates expected from an expert system. The results obtained from the proposed copy-paste fraud detection methods were compared with similar studies in the literature and their advantages were demonstrated.