基于支持向量机的彩色图像水印算法

Color digital watermark based on amplitude modulation and SVR

提出了一种基于回归型支持向量机(SVR)的盲检测彩色图像水印算法。为了改善可见性,根据人眼视觉特性(HVS)自适应选择其嵌入位置。水印嵌入是通过修改图像中像素蓝色通道信息来实现,水印提取时结合图像局部相关性,选取稳定的特征向量并获取SVR训练模型,进而利用SVR训练模型进行预测提取数字水印信息。在嵌入过程中通过高斯模板来均衡调整嵌入水印后对宿主图像影响的强度,仿真实验证明本文算法不仅具有非常好的透明性,而且对诸如叠加噪声、JPEG压缩、平滑滤波、几何变换、图像增强等攻击也具有很好的鲁棒性。

A color image watermarking method with blind detection based on amplitude modulation and support vector regression was presented in this paper. On the basis of the visual masking of human visual system, the embedded positions were chosen adaptively. Basically, the watermark was embedded by modifying the pixel values in the blue channel of an image, while the watermark retrieval was achieved by using a prediction technique based on support vector regression model. The model was trained by learning the relationship between the selected pixel and statistical feature of neighboring area. The technique of Gaussian weighting marks by averaging the luminance of its own pixel and its neighboring pixels was used as an additional scaling factor to tune the strength of embedding watermark. Experimental results show that the presented scheme is invisible and robust against various processing such as cropping, compression, filtering, noise adding, image enhancement and geometrical attacks.