基于结构相似度评价方法的窗口效应

为了提高基于结构相似度图像评价方法的评价效果，对传统的窗口选取准则进行了改进，采用实验对比的方法，通过对窗口改进前后评价效果的对比验证，得到了不同数据库和不同失真类型的最佳窗口。结果表明，传统的单一窗口并不能得到最佳的评价效果，而且现在普遍使用的标准差为1.5的高斯加权窗口的效果也不一定好于普通的方形窗口，不同的失真类型的图像其最优窗口也是不同的。结果表明，对于亮度窗口W1来说，窗口越大效果越好；而对于对比度、结构度窗口W2而言，W2=7时效果最好。该研究成果对优化基于结构相似度评价方法的效果具有重要作用。

Window effect of evaluation method of the structural similarity

In order to improve evaluation effect of structural similarity index image, the traditional window selection criterion was improved. By contrasting the window evaluation value before and after improvement, the best window for different databases and different types of distortion was obtained. The results show that the traditional single window does not get the best evaluation result, and the widely-used Gaussian weighted window with standard deviation of 1. 5 does not better than the ordinary square window. The optimal windows of different types of image distortion are also different. From the experimental results, for luminance window W1 , the bigger the window, the better the evaluation result. And for contrast and structure window W2 , the evaluation result is best when W2 =7. The results play an important role on optimizing the evaluation method of structural similarity index.