Genetic Algorithm Optimized Multi-Feature Fusion Model for Video Homologous Inter-Frame Tampering Detection

Homologous inter-frame tampering forensic methods relying on a single feature has a large number of omissions and false detections for tampered videos where there are sharp movements and complex light changes. To solve these problems, this paper proposes a video inter-frame tampering detection method based on the genetic algorithm-optimized multi-feature fusion model. Firstly, using the video inter-frame continuity, the video grayscale correlation, gradient correlation, and information fidelity correlation are extracted to construct a multi-feature fusion model; then based on the genetic algorithm to optimize the model weight parameter, the change degree of the fused features is calculated as a judgment data point; finally, the change degree anomaly is judged by using Chebyshev's inequality. The experimental results show that this paper's method can effectively improve the detection effect under videos with sharp motion and complicated light changes, etc. Under VIFFD and KTH datasets, the average accuracy rate (precision/%) for the three types of tampering methods is 93.06, 93.61, 91.39, respectively, and the average recall rate (recall/%) is 93.84, 93.445, 92.42, respectively, 92.42. The method in this paper can be better applied in the detection of tampering between frames of the same source of Internet video.