特征提取和特征匹配改进方法的研究  

Approved Algorithm on Feature Extraction and Feature Matching

作　　者：王青松[1,2] 赵西安[2,3] 马超[2] 

机构地区：[1]中国软件与技术服务股份有限公司,北京100081 [2]北京建筑大学测绘与城市空间信息学院,北京100044 [3]现代城市测绘国家测绘地理信息局重点实验室,北京100044

出　　处：《测绘科学技术学报》2014年第4期377-382,共6页Journal of Geomatics Science and Technology

摘　　要：在分析了几种常用的特征提取和图像匹配算法的基础上,针对影像上纹理特征稀少、影像对比度不明显的问题,提出先用高斯差分方法突出图像特征,再用Harris特征点提取算法进行特征提取的方法;针对如同建筑物等相似特征较多的情况,利用双向特征匹配、经过斜率约束得到的结果计算视差,再进行图像匹配,可以将特征匹配限制在相关的小范围内,由此得到的同名点对比较可靠,并且提高了效率。实验证明该算法适合特征稀少不明显区域、建筑物等相似特征较多的图像匹配,具有一定的实际意义。Several feature extraction and image matching algorithms was analyzed in this paper. According to the problem of the feature deficiency or low contrast in some images, Harris algorithm based on Gauss difference image is proposed. Aimed at the similar features such as buildings, parallax can be calculated after two direction matching and slope constraint, then matching is carried on in relevant small size, so more accurate matching points are acquired by matching with parallax. The experiment results show that the algorithm is significant in the images with little features and stereo images with more similar features.

关 键 词：差分Harris特征点 特征匹配 双向匹配 斜率约束 视差 

分 类 号：P237[天文地球—摄影测量与遥感]