基于边缘异常与压缩跟踪的视频抠像篡改检测

摘要

近年来视频的真实性认证成为信息安全领域的一个研究热点。视频的篡改方式多种多样，为了有效判断视频中某个目标的出现是否真实，提出一种视频抠像的篡改检测方法。由于成像设备CFA插值的存在，使得像素存在一定的邻域相关性。为此，提取视频的帧序列，对图像帧进行Sobel边缘检测，计算边缘点4个方向上像素值的偏差来检测边缘点，确定异常边缘点，定位篡改区域，并对篡改区域进行学习，结合压缩跟踪算法快速查找时域上帧的篡改范围，当目标消失跟踪结束之后，在目标出现的最后一帧再次进行异常边缘检测，与最初检测到的异常区域进行对比，查看检测结果是否一致。实验结果表明，该算法大幅提高了检测效率，并且准确率较高。%In recent years, the authenticity of the video becomes a hot research topic in information security field. With various video tamper methods, this paper aims at detecting forgery of video keying to estimate if the appearance of an object is authentic. Because of the existence of the CFA interpolation, pixels have correlation with the adjacent pixels. It extracts video frames, detects the edge of the first video frame by Sobel operator and computes deviation of four directions to judge suspicious edge. And then it studies the tampered area, by using target tracking algorithm follow the area. It improves the detection efficiency greatly and the detection accuracy rate is better.