Digital video frame deletion tampering detection method based on deep neural network

The invention relates to a digital video frame deletion tampering detection method based on a deep neural network. The method comprises the following steps: editing a to-be-detected video into a plurality of video clips; extracting pixel-level differential features and global differential features of adjacent frames from the video clips by using a multi-scale differential convolutional neural network MsDCNN, and fusing the pixel-level differential features and the global differential features to form a multi-scale differential feature; inputting the multi-scale differential feature into an LSTM network using a one-way attention mechanism, obtaining network output scores of the video clips through a full connection layer, and determining whether frame deletion exists in intermediate pointsof the video clips or not; reducing false detection by using a post-processing method based on video local spatial-temporal features. 本发明涉及一种基于深度神经网络的数字视频删帧篡改检测方法。该方法实现过程：将待检测视频剪辑成为若干个视频片段；利用多尺度差分卷积网络MsDCNN从视频片段中提取相邻帧的像素级差分特征