A DEEP LEARNING FRAMEWORK FOR ROBUST DETECTION OF OBJECT-BASED FORGERIES IN VIDEO

Abstract

A crucial component of digital forensics is video forgery detection, which tackles the difficulties caused by video content tampering. In this study, a unique method for detecting video forgeries using Deep Convolutional Neural Networks (DCNN) is presented. Our approach seeks to increase the precision and effectiveness of object-based counterfeit detection in complex video sequences by using deep learning. The suggested strategy introduces novel changes to the DCNN architecture while building on the framework of an established technique that makes use of convolutional neural networks. The network design, training techniques, and data preparation are some of the changes that improve the model's capacity to identify tampered objects in video frames. We explore with sophisticated video encoding standards using the SYSUOBJFORG dataset, which is the biggest object-based fake video dataset to date. When our DCNN-based technique is compared to the current one, it performs better. The findings demonstrate improved resilience and accuracy in identifying object-based video forgeries. This work highlights the promise of deep learning, namely DCNN, in tackling the changing issues of digital video manipulation in addition to making a contribution to the area of video forgery detection. The results pave the way for further investigation into the localisation of fabricated areas and the use of DCNN in video sequences with lower bitrates or resolutions.