融合多特征的视频帧间篡改检测算法

Video inter-frame tampering detection algorithm fusing multiple features

传统的视频帧间被动取证往往依赖单一特征,而这些特征各自适用于某类视频,对其他视频的检测精度较低。针对这种情况,提出一种融合多特征的视频帧间篡改检测算法。该算法首先计算视频的空间信息和时间信息值并对视频进行分组,接着计算视频帧间连续性VQA特征,然后结合SVM–RFE特征递归消除算法对不同特征排序,最后利用顺序前向选择算法和Adaboost二元分类器对排序好的特征进行筛选与融合。实验结果表明,该算法提高了篡改检测精度。

Traditional passive forensics of video inter-frame tampering often relies on single feature.Each of these features is usually suitable for certain types of videos,while has low detection accuracy for other videos.To combine the advantages of these features,a video inter-frame tampering detection algorithm that could fuse multi-features was proposed.The algorithm firstly classified the input video into one group based on its space information and time information values.Then it calculated the VQA features that represented the video inter-frame continuity.These features were sorted by the SVM-RFE feature recursive elimination algorithm.Finally,the sorted features were filtered and fused by the sequential forward selection algorithm and Adaboost binary classifier.Experimental results show that the proposed algorithm could achieve higher tampering detection accuracy.