Comparative Analysis of Convolutional Neural Network and Support Vector Machine Techniques for Video Forgery Detection

Advanced manipulation techniques like the Copy-Move Attack, which includes copying and changing video information, are posing a growing danger to the integrity of digital assets. The validity of digital media depends on the detection of such forgeries, but current techniques frequently find it difficult to strike a compromise between efficiency and accuracy in complex situations. This study tackles this problem by analyzing and contrasting two machine learning methods for video forgery detection: (1) a deep convolutional neural network (CNN) designed for forgery classification, and (2) a support vector machine (SVM) classifier paired with histogram of oriented gradients (HOG) for feature extraction. Both approaches were evaluated on a dedicated dataset of forged and original video frames from the REWIND database, which focuses on Copy-Move manipulations. Experimental results show that the HOG-SVM approach achieved an accuracy of 87.2% and an F1-score of 0.65, while the CNN method outperformed it with 93% accuracy and an F1-score of 0.95. Our contribution lies in presenting a detailed comparative analysis of these methods, highlighting the strengths and limitations of each. These findings underscore the superior capability of CNNs to learn intricate spatial and temporal features, making them a more robust solution for video forgery detection.