基于像素排序的误差扩展方法能实现高保真度的可逆数据隐藏,但嵌入容量较小.针对这一问题,Qu等人提出了以像素为单位的像素排序方法,即使用排序后的上下文像素来预测目标像素.尽管在嵌入容量上实现了较大突破,但Qu算法单一的嵌入策略不能充分利用平滑区域的数据冗余.考虑到以像素为单位的像素排序方法避免了嵌入像素定位问题,提出进一步划分目标像素,并相应地提出了新的基于方向编码的嵌入策略和两轮嵌入式的数据嵌入方案.平滑区域的目标像素预测准确率高,通过双向修改该类像素实施数据嵌入能使得嵌入容量增幅达50％,这意味着后续嵌入操作仅需更少量的非平滑区域目标像素,因此算法可以采用更小的区域复杂度阈值以提高预测准确率.实验结果表明该工作不仅进一步稳固了Qu算法在嵌入容量上的优势,并且有利于提高嵌入图像保真度.与现有基于像素排序的误差扩展方法相比,该算法在多数情况下能取得更优的嵌入图像保真度.%Pixel value ordering prediction based reversible data hiding can obtain high-fidelity marked image at the cost of limited embedding capacity.To achieve larger embedding capacity,Qu et al.proposed the idea of pixel-based pixel value ordering,where each pixel was predicted using its sorted context pixels.Although there is a great breakthrough in embedding capacity enlargement,Qu et al.'s single embedding strategy cannot fully exploit the redundancy of smooth region.In view of the fact that pixel-based pixel value ordering avoided locating the embedded pixel,this paper proposed to divide pixels in more detail and then presented a new exploiting modification direction based embedding strategy and the idea of two rounds data embedding correspondingly.For those target pixels from smooth region which guaranteed high prediction accuracy,the bidirectional modification based data embedding could increase the embedding capacity by 50％.This means that less target pixels from unsmooth region are needed for second round embedding and thus smaller region complexity threshold can be chosen to ensure higher prediction accuracy.Experimental results verify that this work can not only strengthen the advantage of Qu et al.'scheme in embedding capacity,but also help to improve the fidelity of marked image.What is more,the proposed scheme outperforms existing pixel value ordering prediction based schemes for higher fidelity in most cases.
展开▼
