Abstract: In view of digital watermarking algorithms are difficult to deal with some synchronization attacks and composite geometric attacks, this paper proposes a digital watermarking method based on the steady feature area and feature points. It extracts balanced feature points by changing the scale space of corner in Harris-Laplace algorithm. Then, feature direction is decided by gradient factor. Next, it selects invariable feature area according to the steady feature points and feature direction. Finally, watermark is inserted in the feature area. This method can enhance the resistance of the general geometric attacks and composite attacks in data watermarking system. The conducted results demonstrate that the proposed algorithm not only can effectively reply the images that are suffer from geometric attacks, but also avoid the embedded watermark information outside the extracted feature extraction, and it is provided with better robustness.

Key words: digital watermarking, geometric attack, measure space, robustness

摘要： 针对目前数字水印算法对一些去同步攻击和复合性的几何攻击难以应对问题，提出一种基于Harris-Laplacian的特征点均衡化鲁棒数字水印方法。通过改变Harris-Laplace方法中Harris角点的尺度空间来获得分布均匀的特征点；采取梯度指标以决定特征方向，依据特征点与其决定的特征方向以获得不变的特征域；在特征区域中嵌入水印。该方法增强了数字水印对一般性常规几何攻击以及复合性攻击的抵抗能力。仿真结果表明，新方法可以很好地处理几何攻击后的图像改变，同时避免了嵌入的水印信息在提取的特征区域外，具有很好的鲁棒性。

关键词: 数字水印, 几何攻击, 尺度空间, 鲁棒性