中文摘要：

信息隐藏是一种将秘密信息嵌入常见的载体中以实现信息秘密传递的技术.然而,隐写改动都会不可避免地造成各种失真.视频运动矢量隐写则会造成画面质量下降、比特率增加、概率分布改变等变化.这些由隐写造成的影响可能会被已有的或者潜在的隐写分析方法所检测.已有的运动矢量隐写分析特征大致可以分为两类：一类基于局部最优性特征,另一类基于时域和空域相关性特征.为了尽可能降低运动矢量的隐写失真,该文对视频编解码过程中运动矢量相关内容给出形式化描述方法.进一步地,理论分析了关于运动矢量的局部最优性以及相邻相关性两个因素.此外,文中研究了运动矢量隐写造成的最优概率下降的原因,指出了隐写纹理简单宏块的运动矢量不容易造成显著的局部最优性异常.该文讨论了运动矢量分量的分布规律,认为使得分量更靠近分布均值的修改,可以较好保持相关性.通过这些理论研究,提出基于宏块复杂度的视频运动矢量自适应隐写算法（Adaptive Macroblock Complexity,AMC）,其可以分开处理运动矢量局部最优性和相邻相关性两个因素.该算法可以不使用编码实现秘密信息在最低代价路径上的嵌入和提取,而且每个运动矢量可以嵌入2比特秘密信息.算法包括两步：首先,压缩cover视频,不做任何嵌入,记录所有宏块压缩后的复杂度,同时依据分布统计设定一个复杂度阈值.然后,在第2次进行的视频压缩过程中,通过自适应阈值选择低复杂度宏块,将其运动矢量用于LSB（最低有效位）匹配嵌入.对于运动矢量来说,低复杂度宏块较容易保持局部最优,而LSB匹配嵌入可以保持相邻相关性.在实验部分,将不同隐写算法但相同嵌入率的stego样本在不可见性、码率增长、反检测能力等方面分别进行了对比.实验结果表明,cover的PSNR值最高,而AMC排在第二,与之相差不超过0

英文摘要：

g Modern steganography is the art of covert communication which sends secret messages under the regular-looking cover media. However, different distortions will be introduced inevitably by the embedding modification of steganography. The modification of video motion vectors can also result in many distortions, such as poor image quality, bitrate increase and the changes of probability distribution. All of these impacts caused by embedding may be detected by existing or potential steganalysis methods. The established features of motion vector steganalysis can be broadly separated into two categories： one is based on local optimality, and the other is based on spatial and temporal correlation. In order to stay the distortions of motion vector steganography as low as possible, the formal description regarding motion vectors in the process of video coding and decoding is provided in this paper. Moreover, the local optimality and the correlation of adjacent motion vectors have been analyzed theoretically. Furthermore, we study the causes of the local optimal probability decrease due to motion vector modification, and point out that the embedding, in the motion vector of simple texture macroblocks, is not likely to cause significant distortion of local optimality. This paper discusses the distribution rule of motion vector components. The rule reveals that the modification, making the component become more close to the distribution mean value than before, can protect the correlation better. With the aid of theoretical study, we propose a novel adaptive steganography algorithm based on macroblock complexity （ Adaptive Macroblock Complexity, AMC）,which can deal with local optimality and adjacent correlation of motion vectors separately. This algorithm can embed and extract secret information on the minimum distortion path without using any coding method, and every motion vector can be embedded into 2 bits. There are two stages in this algorithm, firstly it compr