中文摘要：

利用运动矢量隐写对相邻运动矢量相似性的扰乱作用,提出一种基于信息熵的视频运动矢量隐分析算法.在视频帧的滑动窗口中选取运动矢量,并分别对水平分量H、垂直分量V、方向D以及长度L等4个系数进行熵值计算以描述相似度.通过移动帧内滑动窗口位置,计算各窗口内运动矢量的熵均值矩阵,构造了16维的运动矢量HVDL特征.为了评估本文HVDL特征的表现,将HVDL与AoSO对比,对不同时期4种嵌入算法的样本进行了隐写分析实验,实验结果表明HVDL对当前安全性更高的隐写算法的检测准确率比AoSO高,且特征运行速度更快.

英文摘要：

A motion vector steganalysis algorithm based on information entropy is proposed by the steganography distortion of the similarity among the adjacent motion vectors. In a video frame, the similarity is measured by the entropy which is calculated from four coefficients of motion vectors in a sliding window： the horizontal component （H）, vertical component （V）, direction （D） and length （L）. Then, the 16-dimension HVDL feature is generated from the mean value of entropy matrix by shifting the position of the sliding window. To evaluate the performance of the HVDL feature in this paper, we make comparisons between HVDL and AoSO by the samples which are generated by different embedding algorithms in different periods. Experiment results show that the HVDL has higher accuracy in detecting for the current safer embedding algorithms and runs faster than AoSO.