Enhance Deepfake Video Detection Through Optical Flow Algorithms-Based CNN

The rapid advancement of deepfake technology has resulted in the widespread distribution of fake audio and videos across network platforms, posing concerns for many countries, societies, individuals, and threatening cybersecurity. Detecting these deceptive videos is a critical challenge in preserving the integrity of information dissemination. We propose a novel approach of employing Optical Flow (OF) algorithms in conjunction with Convolutional Neural Networks (CNN) to enhance deepfake video detection. We are particularly looking into the benefits of combining two separate OF methods, Farneback and Dual TV-L, to achieve optimal results. Our evaluations examine how well each OF algorithm works when combined with CNN-based recognition models. The results show that both OF methods work well independently, but when used together, Farneback and Dual TV-L give us more accurate detection. This study adds to the field by presenting a new way to find deep fake videos by combining these two OF algorithms, which previous literature has not attempted.