Abstract: In wanelet transformation domain, there are lots of semblances among wanelet coefficients at the same direction subbands and spatial location, but different resolutions. This paper proposes an algorithm of robust image watermark based on DWT direction tree structure. The algorithm establishes the model by using support vector machine, hides watermark by adjusting the relation between output and target value of the model. The watermark recovery does not require the original image. Experimental results show that the algorithm possesses good robustness for common image attack, and has lager capacity, better invisibility, better security and super performance.

Key words: image watermark, wavelet transformation, direction tree structure, support vector machine(SVM)

摘要： 在小波变换域中，不同分辨率上、同方向子带内、相同空间位置的小波系数之间有较强的相似性，该文提出了一种基于小波域方向树结构的图像水印算法。该算法利用支持向量机建立小波系数方向树结构模型，通过调整模型的输出值与目标值的大小关系来隐藏水印。提取水印时，不需要原始载体图像。实验结果表明，该算法对常见的图像攻击具有很好的鲁棒性，而且水印嵌入容量大、隐蔽性好、安全性强、性能优越。

关键词: 数字水印, 小波变换, 方向树结构, 支持向量机

CLC Number: 

• TP309