Thesis (Ph.D.) - Cairo University - Faculty of Computers and Artificial Intelligent - Department of Information Technology

Nowadays, there has been an increase in the utilization of digital images. Digital images are used in diverse applications such as in medicine, wars, courts, insurance agencies, electronic media, etc.The aim of this thesis is to develop automatic imaging forgery detection and localization techniques. Two techniques have been proposed to address this issue. The first proposed model is designed to detect copy-move forgery in images. It composes of two stages; the detection stage and the refine detection stage, respectively. The detection stage is performed using Speeded {u2013} Up Robust Feature (SURF) and Binary Robust Invariant Scalable Keypoints (BRISK) for feature detection and in the refine detection stage, non-linear transformation is used to improve detection efficiency.The random sample consensus (RANSAC) technique is applied to remove the remaining false matches from the results. A number of numerical experiments performed using many benchmark datasets such as, the CoMoFoD, MICC-220, MICC-F600 and MICC-F2000 datasets. With the proposed model, an overall average detection accuracy of 95.33% is obtained for the evaluation carried out. Thus; results from different datasets have been established, proving that the proposed model can identify the altered areas, with high reliability. The proposed second model is developed to be a more generalized for the detection of different types of forgery in images by applying a novel modified version of the pre-trained AlexNet model. A novel modified model is proposed to optimize the AlexNet model by using batch normalization instead of local Response normalization, a maxout activation function instead of a rectified linear unit, and a softmax activation function in the last layer to act as a classifier

Issued also as CD