中文摘要：

提出了一种新的基于非下采样Contourlet变换（NSCT）的加权相似性测度计算方法。首先对立体像对中的左右影像分别进行NSCT变换,以得到左右影像的不同尺度、不同方向的高频子带系数;然后利用影像的高频子带系数和RGB三个通道的灰度值,设计了目标窗口和搜索窗口的加权相似性测度计算模型;最后,利用标准立体影像数据,对相似性测度和归一化相关系数等已有测度进行了对比实验。结果表明,NSCT的高频系数的引入增强了相似性测度的稳健性,提高了立体影像匹配的可靠性。

英文摘要：

A new weighted similarity measure based on nonsubsampled contourlet transform is proposed in this paper.In the new algorithm,high frequency sub-band parameters of left and right images in different scales and directions are firstly obtained by respectively carrying out NSCT to left and right images of stereopair.Secondly,according to high frequency sub-band parameters and gray levels in RGB channels of image,the computation model of weighted similarity measure between target window and searching window is designed.Lastly,utilizing standard stereo images,contrast experiments among proposed similarity measure and some known measures such as normalized correlation coefficient,etc..Experimental results show that utilization of high frequency parameters of NSCT enhances the robustness of similarity measure and increases the reliability of stereo image matching.