数字视频篡改检测的被动取证算法研究

发布时间：2018-01-15 07:17

本文关键词：数字视频篡改检测的被动取证算法研究　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：目前数字视频在社会各领域应用广泛,视频逐渐成为一种强大的信息传输媒介。随着计算机技术的发展,各种视频编辑软件迅速普及,普通人无需专业知识即可对获取视频进行有目的地篡改,在一些诸如法庭作证的应用场合,篡改视频可能会导致严重后果,破坏司法公正,影响社会稳定。为了鉴定数字视频的真实性与完整性,数字视频取证技术逐渐成为多媒体信息安全领域中最为重要的研究课题之一,受到越来越多科研人员的关注。在数字视频取证技术中,视频篡改检测研究开展最早,实际应用价值也最为重大。当前很多篡改检测算法一般都需要提取原始帧与篡改帧各自特征进行对比检测,即必须同时获得原始视频和篡改视频,难以进行有效的实际应用。本文基于已有成熟的数字图像取证方法和视频独有特征,针对视频帧内篡改和帧间篡改两种主要的内容篡改方式,聚焦于对待检视频直接盲检测处理,根据提取的相关异常特征分析鉴别视频真伪,以实现被动取证研究。帧内篡改在某一特定的视频帧内部操作,针对视频帧内复制-粘贴篡改操作,本文提出基于边缘提取和特征点检测匹配相结合的被动取证方法。由于原始区域块和复制区域块之间具有很大的相似性,采样视频帧经过预处理之后,本文采用边缘检测算子分析研究视频帧中出现的相同边缘线条。同时,采用SIFT(Scale Invariant Features Transform,尺度不变特征变换)算法检测提取采样帧中的特征点,基于余弦相似性度量方法,提出一种新的特征向量匹配方法以实现特征点的匹配聚类。相比较于2NN(2-Nearest-Neighbor,最近距离和次最近距离比值)方法,本文方法在速度性能上有明显优势且检测准确率更高。实验结果表明本文被动取证方法能有效检测出视频帧中存在的相同区域块,辨别被复制区域的形状和大小,并准确地定位出克隆区域块所在位置。粒度为帧的视频帧插入、删除和复制等帧间篡改操作会改变视频帧的原始位置,本文提出基于时间域相关性分析的视频帧间篡改被动取证方法。采用HSV彩色直方图作为视频帧相似性度量特征,本文分别计算每一视频帧的H-S二维直方图和S-V二维直方图,并进行相邻视频帧的直方图距离比较。根据直方图距离出现的异常变化,本文方法能准确地检测出视频帧插入、删除和复制篡改。基于定位的篡改位置,利用特征相似性匹配,进一步完成了篡改类型的取证复检。视频中存在的运动对象往往是人们重点关注的主体,本文提出基于运动学连续性分析的视频帧间篡改被动取证方法。视频对象的运动学行为由真实运动所决定,但也会被篡改行为所改变。本文采用混合高斯模型背景建模和ViBe(Visual Background Extractor,视觉背景提取)算法对运动目标进行前景提取,通过提取运动目标的异常运动轨迹,分析研究视频帧删除篡改;基于四邻域搜索算法,本文计算运动目标区域的质心坐标,根据视频目标质心参数的异常变化,成功检测视频帧复制篡改。同时,根据运动轨迹和质心参数异常变化出现位置,计算定位出相应的篡改位置。
[Abstract]:At present, digital video widely used in various fields of the society, the video has gradually become a powerful information transmission media. With the development of computer technology, the rapid popularization of video editing software, ordinary people without professional knowledge to get the video to tampering, applications such as testifying in court, video tampering may cause serious consequences, destruction of justice, affect social stability. In order to identify the authenticity and integrity of digital video, digital video forensics technology has gradually become one of the most important research topics in the field of multimedia information security, has attracted more and more attention from researchers. In the digital video forensics technology, video tamper detection research on the earliest the actual application value is the most important. Many current tamper detection algorithms generally need to extract the original frame and the frame of tampering with each characteristic. Than that must be obtained from the original video detection, and tampering with the video at the same time, difficult to carry out effective practical application. This paper has a mature digital image forensics method and video based on the unique characteristics, for intra and inter frame video tampering tampering two main content tampering, focusing on the detected video direct blind detection processing, according to the characteristics of the relevant the anomaly extraction analysis to identify the authenticity of the video, in order to achieve passive forensics. Intraframe tampering in a particular video frame in the video frame for the internal operation, copy paste tampering, this passive forensics method of edge extraction and feature point detection and matching based on combination. Because of the great similarity between the original block copy and block sampling, video frame after pretreatment, this paper uses edge detection operator to analysis the same edge lines appear on the video frame at the same time, Using SIFT (Scale Invariant Features Transform, the scale invariant feature transform) algorithm to detect feature points extraction sampling frames, cosine similarity measure method is proposed based on a new feature vector matching method to achieve the matching feature point clustering. Compared with 2NN (2-Nearest-Neighbor, the nearest and next nearest distance ratio method, this paper) the method has obvious advantages and higher detection accuracy in speed performance. The experimental results show that the passive forensics method can effectively detect the same region exists in the video frame, to identify the replication region of shape and size, and accurately locate the region location. Clone size of video frame insertion, original the location of the deletion and replication of inter frame tampering will change the video frame, this paper take the method of video frames with passive time domain correlation analysis based on using H. SV color histogram as video frame similarity features, the thesis calculates each video frame H-S two-dimensional histogram and S-V histogram comparison, histogram distance and adjacent video frames. According to the abnormal changes of the distance histogram, this method can accurately detect the video frame insert, delete and copy location tampering tampering. Based on the position, using feature similarity matching, completing a further tampering with the type of evidence review. Moving objects in the video there is often the main focus of attention, this passive forensics method tampering with the video frame continuity analysis of kinematics based on kinematics behavior. The video object is determined by the real movement, but also will be behavior with change. This paper uses the Gauss mixture model background modeling and ViBe (Visual Background Extractor, visual background extraction algorithm of moving target) Standard for foreground extraction, the abnormal trajectory of moving object extraction, analysis and research of video frame deletion forgery; search algorithm based on four neighborhood, this paper calculates the centroid coordinates of target region, according to the abnormal change of the video target centroid parameters, the successful detection of video frame copy tampering. At the same time, according to the motion trajectory and centroid position parameter changes the calculation, to locate the appropriate tampering position.

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】