A new deep learning-based method to detection of copy-move forgery in digital images

Thanks to ease of using image manipulation programs, copy move forgery is the easiest image modification approach, which aims to duplicate or remove objects in the image. The methods to detect this type of forgeries are partitioned into; block-based, keypoint-based approach that use hand-crafted features. A new deep learning-based forgery detection scheme is presented. An existing trained model AlexNet is utilized to extract feature vectors of image overlapped subblocks. After obtaining features, the similarity between feature vectors has been investigated for the detection and localization of forgery. The method has higher accuracy rate than the considered traditional method in the literature as reported in the obtained test results.