DEVELOPMENTS OF ALGORITHMS FOR THE DETECTION OF FORGERY IN VIDEO SIGNAL

Authors:

Seema S. Gharge¹, Priyaramani P. Kale², Shiveshnu K. Shinde³, Siddhati S.Awalekar⁴

Prof. A.S.Mohite Department & College

Kolhapur Institute of Technology’s College of Engineering, Kolhapur

---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract - With the rapid growth of artificial intelligence and deep learning technology, the creation of realistic modified videos often known as deepfakes has become increasingly complex and accessible.These forged videos pose serious threats to digital media authenticity, public trust, cybersecurity, and legal evidence integrity. This paper proposes a deep learning-based algorithm for the detection of video forgery using the EfficientNet-B0 architecture. The system extracts uniformly spaced frames from input video files, preprocesses each frame through resizing, normalization, and tensor conversion, and subsequently processes them through a pretrained convolutional neural network. Frame-level predictions are aggregated using a majority voting mechanism to arrive at a final binary classification: AI-Generated or Normal Video. A confidence score is generated to quantify the certainty of the prediction. The proposed system is implemented with a graphical user interface (GUI) that supports multiple video formats and provides real-time result visualization. The experimental evaluation showed an overall accuracy of 95.3% on test datasets and outperforms the existing methods such as ResNet-50 and XceptionNet. The proposed system provides a reliable, computationally efficient, and user-friendly solution for automatic video forgery detection.

Video forgery, Deepfake detection, EfficientNet-B0, Frame extraction, Majority voting, Deep learning, Digital media authentication