Robust Copy-Move Forgery Detection Technique Against Image Degradation and Geometric Distortion Attacks

The rapid pace of the digital age has led to an increase in the illegal copying, reproduction, and creation of forgeries of digital images. Copy-move forgery is one of the most common image forgery techniques which is used for tampering with image content. In this paper, a novel scheme is proposed to detect copy-move forgery and if the image is forged, it is aimed to reveal duplicated regions. This study is based on the combination of a keypoint-based method that fails especially when a high rate of blurring attack but is successful in geometric transformation attacks, and a segmentation-based method proposed to gain resistance to blurring attack. The method firstly queries the presence of a blurring attack in the suspicious image. In case of the presence of this attack, forgery detection is made with the segmentation-based module and in the other case with the keypoint-based module. In the keypoint-based module, in order to minimize the effect of possible noise addition attacks applied to the image, a denoising step is performed with the deep neural network. After that in this module, it is proposed to use AKAZE keypoints to reveal duplicated regions, and in the tamper localization step, the Ciratefi-based approach is applied. In the segmentation-based step, it is proposed to segment input images with two-layered segmentation with entropy-based and color-based segmentation. Then, DCT based features are extracted from the image sub-blocks in the segments, the blocks with the same segment label are matched with the feature vectors obtained among themselves. The experimental results demonstrate that the proposed method has an overall performance that is superior to popular approaches on two open-access copy-move forgery datasets.