基于运动估计的视频隐写检测算法

提出一种基于运动估计的的视频隐写检测算法。通过块均方误差的变化研究信息嵌入对运动估计的影响，发现运动矢量对隐写具有敏感性，且分块越小，对隐写越敏感。反映视频时变特性的运动矢量场被作为隐写检测中视频数据的表征，特征提取过程中，先取定分块大小，使用最小均方误差块匹配运动估计算法，得到运动矢量场。再提取运动矢量场3个方向相邻元素的共生矩阵，使用共生矩阵的主对角线及其相邻元素作为特征。使用支持向量机分类器实施检测，实验结果表明文中算法与Budhia的算法相比，具有更好的检测性能。

Digital Video Steganalysis Algorithm Based on Motion Estimation

A motion estimation based video steganalysis algorithm is proposed. The effect of information embedding on video motion estimation is investigated by the change of block mean square errors. The motion vectors are sensitive to information embedding, and the sensitivity increases with the decrease of the block size. The motion vector fields which reflect the time-variable characteristic of videos are used as the representation of video data in detecting stego-video. Firstly, a block size is given, and the motion vector fields are obtained by the minimum mean square error block matched motion estimation algorithm. Then the co-occurrence matrixes of three directional adjacent elements in the motions vector fields are calculated. The main diagonal elements and their neighbors of the co-occurrence matrixes are selected as the features. The support vector machine is adopted as the classifier. The experimental results show that the performance of the proposed algorithm is better compared with that of Budhia algorithm.