Image forgery detection in forensic science using optimization based deep learning models

Image forgery detection has become a hot research topic in security and forensics applications. Keypoints (KPs) based image forgery detection extracts image KPs and utilizes local visual features for identifying forgery regions which shows excellent performance on the basis of memory and robustness over attacks. But, these methods are lacking in handling the cases when forgeries in small regions, in which KPs are limited. To address these challenges, this work introduces a fast and efficient forgery detection using the processes like pre-processing, feature extraction and optimal classification. Initially, the RGB image is converted into L*a*b color space. Then, the KPs are determined using the Eigenvalue Asymmetry (EAS) and after determining the KPs, the texture features like Haralick and skewness are extracted. Then, these features are classified by the deep learning (DL) classifier Deep belief network (DBN) with an adaptive crow search algorithm (ACSA). The proposed DBN-ACSA classifier categorizes the detected image as a normal or forgery image. The experimental results are carried out on the Vision, UCID and CoMoFoD datasets and achieved accuracy, precision, F1 score, recall and ROC values.