基于帧间预测误差扩展的可逆视频水印

Reversible Video Watermarking Based on Inter-prediction Error Expansion

为了提高可逆视频水印的嵌入容量和嵌水印视频质量,提出一种使用运动估计和预测误差直方图修改的算法.在嵌入端,根据视频中相邻帧间的内容关系,对每一帧使用运动估计获得其像素的预测误差,生成预测误差直方图,并通过扩展位于直方图峰值点的预测误差来嵌入水印;在每帧的嵌入过程中会产生少量用于提取水印和还原视频的头信息,包括运动向量和边界表等,将其与水印一起嵌入到该帧的参考帧中.在提取端,先获取参考帧中的水印和头信息并还原参考帧,保证提取操作具有正确的上下文,然后根据获取的头信息和还原后的参考帧提取当前帧中的水印.实验结果表明,采用文中算法获得的预测误差直方图具有高度的集中性,并且对像素值的修改十分微小,比其他可逆水印算法具有更大的嵌入容量和更好的嵌水印视频质量.

To increase the embedding capacity and to improve watermarked video quality of reversible video watermarking,a novel algorithm using motion estimation and modification of prediction error histogram is proposed.On the embedding side,prediction errors of each frame are obtained by motion estimation considering the relationship among adjacent frames of a video;histogram of prediction errors is generated,and the prediction errors at the peak points of the histogram are expanded to hide data.A little head information to be used in the watermark extracting process,including motion vectors,boundary map,etc.,is appended by the watermark and embedded into the reference frame.On the extracting side,the watermark and head information hidden in the reference frame are retrieved and the reference frame is recovered first,which guarantees a correct context for the extraction of the current frame;then the hidden data in the current frame is extracted according to the head information as well as the recovered reference frame.Experimental results have demonstrated that the proposed algorithm can gain a highly concentrated prediction error histogram and provide a much larger embedding capacity and a higher quality of watermarked video than other reversible watermarking algorithms.