Video Forgery Detection Using Distance-based Features and Deep Convolutional Neural Network

Technological advancement of image and video processing tools made the tampering of videos faster and easier, and even a non-professional can make it simpler. Video tampering becomes a serious problem because it promotes fake news in social media and fake videos manifest in the court. Hence, the identification of forgery done in video is a difficult task. This paper introduces video forgery detection through Deep Convolutional Neural Network (DeepCNN). Initially, from the input video, the keyframe extraction is done using discrete cosine transform (DCT) and Euclidean distance. Then, the face objects are detected from the extracted keyframes using the Viola-Jones algorithm. After that, the light coefficients are calculated from the detected face object using 3D shape and texture models. Then, the features of the 3D face are extracted by employing distance measures. Finally, for video forgery detection the input video and the distance-based features are given to the DeepCNN. The devised method achieved maximal accuracy, True Positive Rate (TPR), True Negative Rate (TNR), and Receiver Operating Characteristics (ROC) of 91.02%, 83.64%, 94.89%, and 95.56%, respectively.