An Secure and Effective Copy Move Detection Based on Pretrained Model
Baraa Tareq Hammad, Ismail Taha Ahmed, Norziana Jamil
Abstract
The amount of counterfeit or faked photographs that depict inaccurate or incorrect information has increased. Hence, the Digital image forgery has become a serious problem. Copy move forgery is more risky since it involves copying and pasting a portion of an image into another region of the same image to hide information. Conventional Copy Move Forgery Detection (C-MFD) approaches have limitations in terms of performance. The explanation for this is that the discriminative ability and partially invariant to specific transformations of manual-crafted features are insufficient. Therefore, in this paper we used the AlexNet deep learning model to extract the image features and applies the ReliefF feature selection algorithm to get effective features. The logistic classifier is then given selected features to determine whether the image is forged or not. On the publicly accessible benchmark datasets MICC-F600 and MICC-F2000, the proposed method is tested. The precision rate of the presented method based on AlexNet model is equal to 94 %.