An Efficient Approach to Detecting a Copy-Move Forgery in Videos

In the past years, the advancement in editing technologies, especially multimedia, has contributed to increased image and video modification use. However, object-based video tampering, such as including or excluding objects in the video frames, poses a major challenge to video authentication. Since multimedia content is commonly used across numerous fields, video forgery detection is critical to maintaining media integrity. Regarding different types of manipulations in the domain of videos, copy-move forgery is typical and rather difficult at the same time. This study introduces a relatively efficient DCNN model to detect forged copy-move videos. The proposed method can be described as follows: first, the video is divided into frames, and then a convolutional neural network model is employed to traverse each frame to establish its features. We then use the described features to train a new CNN model, making it possible for us to determine whether a particular frame is real or fake. Also, the structure incorporates batch normalization to enable easy layer weight initialization, ease in training at a higher learning rate, high accuracy, non-overfitting, and process stability. We also conducted extensive practical experiments on a massive dataset of videos, which included both original and manipulated content. We use specific performance metrics like accuracy, precision, recall, F1-score, and the Matthews Correlation Coefficient to assess the performance of the suggested model. The recommended method demonstrated superior performance compared to all previously proposed methods on the GRIP, VTD, and SULFA datasets. The model’s accuracy was 95.60%, 96.70%, and 100%, with the shortest time of 25.10 sec, 27.35 sec, and 20.22 sec, respectively.