Abstract: Conventional SNR and PSNR can’t evaluate watermarked image quality well.Human vision has different sensitivity threshold to different intensities of luminance and different degrees of texture in image.SNR and PSNR do not distinguish the differences between them.By calculating the JND-value（Just Noticeable Difference） for each DCT-coefficient of image in DCT domain with Watson’s DCT-domain JND model，we assign different weight for each component of image and get an optimized quality-evaluation model.Comparison is implemented between the conventional model and the optimized one by different experiments.The result shows that the new model outperforms the conventional measures of SNR and PSNR.The quality of image evaluated by the new model is more close to the human visual sensation.The new model is applicable to different kinds of image and different watermarking methods.And it can be used to evaluate the imperceptibility performance of the watermarking method.

摘要： 传统的SNR和PSNR不能很好地评价水印图像的质量。人类视觉对图像中不同强度的亮度区域和不同程度的纹理区域具有不同的视觉阈值，SNR和PSNR没有对这些不同进行区分，利用Watson提出的DCT域JND（Just Noticeable Difference，刚辨差）计算模型计算出图像中每个DCT系数的JND值，为图像中不同成分设置不同的权值，得到了优化的图像质量评价模型。用不同方法对传统模型和本文模型进行对比实验，结果表明新模型优于传统模型，其评价结果更接近人的视感觉，适用于各种不同的灰度图像和水印算法，可以对水印算法进行不可感知性性能评价。