Two-step intelligent approach for photo image fragment forgery detection and identification

Abstract

The paper presents a two-step approach for detection and identification of fake (forged) fragments in photographic images. At the first step, the image is assigned to an original or a fake class by sequentially analyzing the image metadata (EXIF), then performing the error level analysis (ELA), and finally performing neural network analysis of the image. At the second step, the fake image is divided into fragments, each one is analyzed using a neural network for the fake detection, then the fragments that had recognized as a fake are analyzed by a classifying neural network, which makes it possible to identify specific types of fakes. Software has been developed that implements the proposed approach. The training of the used neural networks was carried out on the CASIA V2 dataset, which contains 4795 images (1701 genuine and 3274 fakes). Studies have been carried out to verify the effectiveness, for which, using the developed software 60 photo images were processed, 30 of which are genuine, and 30 are fakes. The developed software correctly identified both the fake fragments and determined their types.