一种基于线性神经网络的半易损水印技术

本文提出了一种基于线性神经网络的半易损水印技术。该算法利用线性神经网络对小波低频系数及其邻域系数的关系进行建模，然后利用被选的嵌入点处的小波系数与模型输出值之间的大小关系来嵌入二值水印。基于该模型提取水印时不需要原始图像和水印，只需模型的参数即可。实验表明，该算法不仅对于恶意篡改有较强的定位能力，而且可
可以抵抗JPEG压缩。

A Linear-Neural-Network-Based Semi-Fragile Watermarking Technique

A linear-neural-network-based semi-fragile watermarking technique is proposed in the paper. The method establishes a relational model among these coefficients and other coefficients in their vicinities in the low frequent band using linear neural networks. Based on the linear neural network model, watermarks can be extracted only with the parameters of the model without the participation of the original image and watermarks. Experimental results show the proposed method can not only locate the tempered areas exactly but also resist the JPEG compression.