Identifying Forged Images Using Image Metadata

摘要

Nowadays, we receive, send and click a number of images frequently. Multimedia messages especially images can be easily transferred to anyone in no time and no cost just like text messages. There are many software easily available that can modify an image. With such a wide usage of images in communication networks it becomes difficult to conclude if an image is an original one or a forged one. This paper proposes the usage of image metadata, which is much smaller in size as compared to contents of an image, to authenticate an image. Further, the architecture of a system using a neural network is proposed which extracts attributes from the metadata of an image to predict if an image is an original one or a forged one. The system is also capable of predicting the operation applied and the software/tool used to modify the original image in case of a forged image. Experiments have been conducted on JPEG images which are modified with operations like Crop, Rotate, Resize, Compress, Compress_Rotate and Crop_Rotate to validate our system.