Enhancing Video Tampering Detection Using Dynamic Temporal LSTM With Adaptive CNN

Abstract

In the domain of information technology, video tampering detection has become hyper critical principally with the increase in deep fake as everyone is having affordable access to the internet. The long established methods lack in detecting the manipulated content specifically for temporal disordered and variant frames. In order to overcome such issues, the suggested innovative method encompasses Dynamic Temporal Warping (DTW) within the LSTM framework to efficiently focus on these temporal misalignments, which are usually experienced in real-world scenarios. Hence, an adaptive CNN component is introduced to dynamically adjust for frame rate variations, significantly reducing misclassification rates. Moreover, the proposed method is implemented in Python and it outperforms existing approaches, achieving 96.83 accuracy, 96.9 precision, 96.9 recall, 97.3 F1-score and 98% sensitivity, while also maintaining a lower false positive rate of 2%, making it highly effective for real-time tampering detection in deep fake applications.