Frame-Deletion Detection for Static-Background Video Based on Multi-scale Mutual Information

摘要

Due to enormous free video editing software on the Internet, tampering of digital videos has become very easy. Authenticating the integrity of videos and detecting any video forgery is a big challenge to researchers. In this paper, an algorithm based on the normalized mutual information feature is proposed to detect the frame-deleting videos which are hardly identified by human visual. The proposed method is composed of two parts: feature extraction and abnormal point detection. Firstly, based on information theory, the normalized mutual information is defined on the single scale visual content of adjacent frames. After using the Gaussian pyramid transform on every frame, the description operator of multi-scale normalized mutual information is computed by linear combination. In the stage of discontinuity point detection, video forgery is identified and the tampering point is localized by performing modified generalized ESD test.