一种结合交叉熵和投影特征的图像匹配算法  

Image Matching Algorithm Based on Cross-entropy and Projection Features

作　　者：周军妮[1] 杨润玲[1] 王燕妮[1] 江莉[1] 

机构地区：[1]西安建筑科技大学信息与控制工程学院,西安710055

出　　处：《小型微型计算机系统》2013年第2期405-408,共4页Journal of Chinese Computer Systems

基　　金：陕西省教育厅专项科研基金项目(11JK1023)资助

摘　　要：基于图像交叉熵的图像匹配方法对于噪声不敏感,并且具有一定的抗几何失真能力,但算法复杂度高,不适合用于实时匹配系统中.而投影变换可将图像的二维灰度降为一维的特征向量,且还具有抗噪性好的特性,因此定义图像的局部交叉投影熵,提出了一种新的图像匹配算法.该算法首先计算模板图的行、列投影;然后计算模板图和实时图的交叉投影熵;最后根据行、列交叉投影矩阵确定出最优匹配坐标.新算法不仅具有较好的抗噪和抗几何失真性能,并且提高了在强光照射及云层遮挡情况下的匹配能力.通过实验仿真并对比局部熵、局部投影熵、局部交叉熵和局部交叉投影熵四种算法的匹配效果,表明该算法不仅匹配效果良好,并且计算速度快,是一种精确而实用的图像匹配方法.Image cross-entropy matching algorithm has neither noise sensitivity nor rotational variability, but it is not suitable for real- time matching system because of its high complexity. Because the feature dimension of an image can be reduced by projection trans- formation, cross projection entropy is defined and a novel image matching algorithm is proposed. Firstly, row and column projections of the template image are separately calculated; secondly, the cross projection entropy between the template image and the real-time image is computed; lastly, the optimal matching coordinate is determined according to the cross projection entropy matrix. This image matching algorithm has good anti-noise capability, and good matching results can be got in high light condition and in cloud cover condition. A contrast experiment of four image matching methods of local entropy, local projection entropy, local cross-entropy and local cross projection entropy is conducted. The results show that the proposed algorithm has fairly good matching performance, high running speed. It proves to be a precise and practical image matching method.

关 键 词：图像匹配 局部熵 局部投影熵 局部交叉熵 局部交叉投影熵 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]