【作者】 胡学龙； 邓小颖； 王斌；

【Author】 HU Xue-long1,2+, DENG Xiao-ying1, WANG Bin1 1(Department of Electronics and Communication, Coll of Inf Engin, Yangzhou University, 225009, China) 2(Jiangsu Prov Key Lab for Compt Inf Process Tech, Suzhou University, Suzhou, 215006, China)

【机构】 扬州大学信息工程学院电子与通信工程系；

【摘要】 用于版权认证的水印只需要判断它是否存在。本文首先分析了加法和乘法两种嵌入模型下检测器的理论输出分布、影响漏检率和虚警率的因素,进而提出了改进的嵌入方法。该方法根据水印与特定图像的相关系数采用不同的嵌入策略,使检测结果分布在正负轴两边,根据其绝对值可以判断水印是否存在。最后,讨论了有噪声情况下算法的稳健性。更多还原

【Abstract】 We only need to judge whether the watermarking exist, when the watermarking is used for copyright certification. The information given by this kind of watermarking only one bit. So false alarm rate and omission factor is the most important criterion for judging the watermarking. In this paper, the probability distribution of the detection is analyzed under two kinds of embedding model--addition model and multiplication model, and get the factors effecting the false alarm rate and omission factor. To gauss and twovalued watermarking signal, the embedding watermarking vector is wished to have big direct component, high dimensions and little variance. The watermarking intensity is wished to be high, but the invisibility of the watermarking must be considered together. The examples given by the paper of embedding watermarking in the spatial domain and DCT direct coefficient proves the result above. According to the correlation coefficient of watermarking and embedding watermarking vector, new embedding method is proposed in this paper, make the detection coefficient exist in the positive and negative horizontal axis, and estimate the watermarking according to their absolute values. For addition model,if the correlation coefficient of watermarking and embedding watermarking vector is positive, the watermarking is added to the vector, else, if the coefficient is negative, the watermarking is subtract form the vector. For multiplication model, we can not make the detection coefficient exist in the two sides of horizontal absolutely, but the experiments prove that it have little effect to false alarm rate and omission factor. The robustness of our method is discussed under the effecting of noise. According to the theoretically and experimental result, the probability distribution of the detection can be considered as a product of a Gaussian function and an error function. The product function curve, in the side where might product errors, is steep. So the method of the paper can ensure very little false alarm rate and omission factor, though the noise is big. The high relativity of the noise from intended attack and embedding watermarking vector should be prevented. In this paper, embedding the watermarking in the DCT coefficient is better than in the spatial domain. Above all, The method proposed by the paper is effective and applied.更多还原