Document Image Forgery Detection Based on Deep Learning Models

With the improvement of the communication speed and the popularization of the Internet, images have become the most common information medium in life. At the same time, the adverse effects of forged images in the media, credit investigation, finance and academic fields are becoming more and more significant. Therefore, in recent years, the research on forged image identification algorithms has been active worldwide. Image forgery has different classification methods. According to whether the forgery uses deep learning methods, it can be divided into deep forged images and traditional forged images. It can also be divided into ordinary image forged and document image forged according to whether the image is a text image. Different forgery methods will leave different forgery traces in the image, corresponding to different forgery identification methods. Aiming at document forgery images, this paper proposes a forgery detection algorithm based on deep learning and fusion of error level analysis (ELA) information. Compared with the previous forgery identification algorithms, the algorithm in this paper can not only identify whether the document image is forged, but can also locate the forged text area. The algorithm proposed in this paper supports the detection of document image forgery generated by cutting, copying, erasing and deep learning methods. The detection algorithm of this paper participated in the fifth forgery detection competition of Ali Tianchi and won the 32nd place among 1470 participating teams.