Object based Forgery Detection in Surveillance Videos using Optimized CNN

Abstract

The mass consumption of digital multimedia content is the most accepted and user-friendly presently. This digital interactive environment is based upon images, text, video, audio, etc. The advancement in technology also raises the risk of the fabrication of original content and generating fake content. A video is considered the most trust-able source for an individual, but advanced tools can easily forge the video content. To maintain the authenticity of the video, a scientific approach is required. In this paper, we proposed a pixel-based motion residual technique to uncover the traces of forgery in surveillance videos, followed by an optimized CNN with 92% accuracy on testing. The novelty of proposed network is the optimization in terms of trainable parameters are reduced, network’s complexity is very much reduced, fast operating and computationally cheap without compromising the accuracy.