基于帧间相关性的盲视频数字水印算法

Blind video watermarking algorithm based on correlations between neighboring frames

提出了一种在原始视频中嵌入水印的算法。嵌入的水印由2部分组成:标志水印和信息水印。首先从一段原始视频里随机选出若干视频帧,在这些帧的像素平面亮度分量上固定位置嵌入标志水印,再根据密钥选取其他若干图像块在其离散余弦变换(DCT)直流系数中嵌入信息水印。然后在整段视频的每一小段中重复嵌入水印。提取水印时,只需从整段视频中截取足够包含完整信息水印的一小段视频,通过搜索标志水印快速地选出包含水印信息的帧。然后利用水印帧与前后相邻视频帧之间内容的高度相关性,得到嵌入水印位置像素值的估计,与包含水印信息帧相应位置做比较,提取出水印信息,实现了水印的盲提取。实验结果表明本算法能够抵抗压缩、缩放、帧删除/插入等各种攻击,具有较强的健壮性。

A blind video watermarking method was proposed, in which the watermark composed of symbol part and information part was embedded into the original video frames. After some video frames hiding the watermark were selected randomly from one segment of the video sequences, the symbol water mark was embedded into low frequency discrete cosine transform （DCT） coefficients of fixed blocks of such frames, whereas the information water mark was embedded into direct current coefficients of some other blocks selected by a cryptographic key. Repeat embedding till the whole video was watermarked. To extract the information watermark, the frames holding watermark were found out by detecting water the symbol mark rapidly from one video segment whose length was long enough to hold the whole watermark. Then, by using the high correlations between the watermarked frame and its neighboring frames, the pixel values of the blocks where the watermark was embedded can be well estimated. Comparing the estimated values with the current values in the corresponding positions, the information watermark could be reconstructed. Experimental results show that this video watermarking method is robust against attacks such as compression, scaling, frame dropping, etc.