Deep Learning based Blind Source Identification of whatsApp and Facebook Images

Source Identification is the forensic problem of source device detection of digital images/videos/audio. In this paper, we deal with the problem of Source Camera Identification (SCI) of digital images. The existing forensic techniques of image source identification fail in source identification of images taken from Social Networking platforms, mainly due to the inherent compression characteristics induced in to those images. In this work, deep neural network based architecture is proposed to identify the source of Facebook and whatsApp images. We used InceptionV3 model, which is a 48-layers deep. It surpasses the state-of-the-art performance in terms of online source networks (OSN) image source identification accuracy. We considered smart phone devices in this work for our experiments, and present our results on the recently published VISION dataset.