Abstract: In image watermarking system，a satisfying result can be obtained where JND threshold is used to control the embedding location and intensity and determine the extraction location in blind algorithm.The robustness of JND plays a key role in blind watermark detection.Through the study of whole watermarking system，a novel JND computation method is proposed，which enhances the robustness of JND threshold.Most of the embedded watermark can be effectively located even when the image is significantly altered，thus significantly improve the detection result in blind extraction.At the same time，a novel blind detection algorithm called matching detection based on statistical characteristics is given.Matching detection enhances detecting responses and takes the probability of false alarm into consideration，thereby obtaining high credibility.

Key words: digital watermarking, wavelet transform, Just Noticeable Difference（JND） threshold, matching detection

摘要： 在图像数字水印技术中，采用JND阈值对水印的嵌入位置与强度进行控制，并在盲检测中作为搜索水印嵌入位置的依据，能够得到较理想的效果。JND阈值的鲁棒性对理想的盲检测结果至关重要。通过对数字水印系统的研究，提出了基于图像低频小波系数的定量纹理测度的JND阈值计算模型。该模型明显地提高了JND门限值的鲁棒性，即使在图像受到较大幅度改动的情况下，仍能很好地定位嵌入水印，从而有效地改善了盲检测的效果。同时给出了一种基于随机统计特征的匹配检测算法，该算法有效地提高了响应值，同时考虑响应值对应的虚警概率，从而使检测结果具有较强的说服力。

关键词: 数字水印, 小波变换, JND阈值, 匹配检测