VIDEO FORGERY DETECTIONUSING DNN(DL)

Abstract

Video forgery has become increasingly common due to advancements in editing tools and the widespread distribution of digital media. Detecting manipulated video content is essential for ensuring authenticity in surveillance, journalism, entertainment, and legal investigations. Traditional video forensic techniques rely heavily on handcrafted features, which often fail against modern, complex manipulations. This paper presents a Deep Neural Network (DNN)-based framework for detecting video forgery by analyzing spatial and temporal inconsistencies in video frames. The proposed approach extracts deep features using convolutional layers and models temporal dependencies using sequential learning layers. The system is evaluated on benchmark video forgery datasets containing frame duplication, frame-deletion, splicing, and copy–move manipulations. Experimental results demonstrate that the DNN outperforms conventional models in terms of accuracy, generalization, and detection robustness, achieving 94–96% accuracy depending on the manipulation type. Confusion matrix analysis confirms minimal false positives, proving its reliability in real-world scenarios. The proposed system provides a scalable and automated solution for modern video forensic applications.