中文摘要：

讨论一类基于奇异值分解（SVD）的水印算法的高虚警率问题及其产生原因。分析认为SVD水印算法虚警率高的根本原因是：图像SVD分解的基空间与图像内容相关;奇异值向量与图像之间并不存在一一对应关系,不能刻画图像的几何结构。算法主要缺陷并不是出现在提取算法,而是嵌入算法仅仅植入了水印图像的奇异值向量,没有水印图像在基空间的结构信息。因此,该文提出了一个基于分块SVD和离散余弦变换（DCT）分解的改进算法。该算法通过在版权图像的奇异值向量中嵌入水印图像的DCT系数,克服了经典SVD水印算法的虚警问题,因而稳健性高。...

英文摘要：

The problem of high false alarm for a class of singular value decomposition（SVD） based image watermarking algorithms is investigated and the corresponding essential reasons are discussed.The intrinsic reasons that cause the high false alarm problem are： the basis space of SVD is relative to image content;there is no one-to-one correspondence between singular value vector and image;singular value vectors have no information of the structure of image,thus the most important reason is a result of false conception to insert watermark singular value vectors without using the structure information of watermark.An improved image watermarking algorithm is proposed combining the block-SVD and discrete cosine transform（DCT） decomposition.The DCT coefficients of watermark image are embedded into the singular value vectors of host image,and the improved algorithm can overcome the high false alarm problem which exists in the classical SVD based algorithms and also it has good robustness.Experimental results prove the effectiveness of the theoretical analysis and the improved algorithm.