Abstract: This paper presents a new high robust watermarking algoritfm for digital image. It is found that the differences of mean values of two near blocks in one image are distributed regularly by experiments, and most of them are between an upper bound and a lower one. Hence, anexpression for generating the deviation can be formulated based on the binary watermarks, and the binary watermarks are embedded accordingly bymodifying the gray values of the corresponding blocks in spatial. Watermark detection is implemented by comparing the new difference of mean values of watermarked image blocks with the threshold values. The experimental results demonstrate that the proposed algorithm is robust to many attacks, such as JPEG, noise and filtering etc. This watermarking detection has not reference to the original image or watermarking image, and it realizes effectively the blind detection of original watermarking.

Key words: Difference of mean values; Binary watermarking; Blind detection

摘要： 提出了一种新的鲁棒性高的数字图像水印算法。通过观察得知，图像的两个相邻分块的亮度值的均差具有一定的分布规律，即大多数值落在某个下限和上限之间。由此，结合二值水印的值构造一个偏差公式，再用此公式去修改分块的值以嵌入二值水印图像。在水印提取时，计算分块的新的均差值并与所设门限进行比较，从而检测嵌入的原始水印值。实验结果表明，该算法对JPEG 压缩、各种噪声和滤波等攻击具有较强的鲁棒性。该水印检测算法没有参考原始图像或原始水印图像，有效实现了原始水印的盲检测。

关键词: 均差；二值水印；盲检测