Abstract: To seek the good tradeoff between robustness and unobtrusiveness, the cover data is divided into nonoverlapping blocks and an independent distortion situation is assigned to each block, which is called localized quality evaluation. Two specific implementations applying the Signal to Noise Ratio and Watson's visual model as the quality metrics are respectively achieved in the sense of maximizing the detection value. And the corresponding upper limits are derived under the given distortion constraints. Then, some practical adjustment strategies of those relative parameters are provided. The approach also introduces an improved linear correlation detector using side information and analyzes its performance mathematically. A lot of experimental results show the presented schemes obtain very significant performance improvement in terms of image transparency and robustness.

Key words: localized perceptual quality evaluation, imperceptibility, robustness, spread spectrum watermarking

摘要： 为获得鲁棒性与不可感知性之间的最佳折中效果，将载体信号分成多个信号块，并为每个信号块规定一个独立的失真条件，称为局部质量评价。在检测值最大化的意义下，分别以信噪比和Watson视觉模型描述失真条件探讨了两种具体的水印实现方案，并且在给定的失真条件下推出了相应的检测值上限，然后对相关参数提供了一些实际而有效的调整策略。该方案中还引入了一种改进的线性相关检测器。它利用了载体信号的边带信息提升检测值，并从理论上对其性能进行了分析。大量实验结果表明，该方案在水印不可感知性和鲁棒性两方面均获得了非常显著的提升。

关键词: 局部质量评价, 不可感知性, 鲁棒性, 扩频水印