An Enhanced Video Forgery Detection Using <scp>BiLSTM</scp> With <scp>HOG</scp> and <scp>LBP</scp> Features

ABSTRACT Videos are very important medium to share news and information and for legal proof. So the authentication of video is a very sensitive topic in video forensics research. Modifications in videos using different editing tools to change the actual information of the video intentionally is called forgery. In this study, we suggest an approach for detecting inter‐frame video forgeries, utilizing a fusion of HOG (histogram of oriented gradients) and LBP (local binary pattern) feature extractors, along with BiLSTM (bidirectional long short‐term memory) classifier for temporal learning in sequence. The suggested framework is assessed on Kaggle's video forgery dataset and the VIFFD dataset, employing cross‐validation to ensure robustness performance estimation. Experimental result demonstrate the effectiveness of the approach across different forgery types, with the higher accuracy of 97.08% achieved for deletion‐type forgery in Kaggle's dataset. These findings suggest that integrating handcrafted features with sequential deep learning models provides a promising and lightweight solution for reliable inter‐frame forgery detection in practical forensic applications.