摘要
对于尺度不变特征变换方法(SIFT)应用的图像拼接过程中存在错误匹配的问题,结合拓扑学理论,提出一种去除错误匹配的SIFT改进算法。定义特征点间的拓扑关系,对于两幅图像的特征点间拓扑关系做异或运算得到判断矩阵,获得拓扑关系完全相同的特征点集,有效提高特征点的匹配精度。另外,提出利用特征点重复率、特征点辨识率以及图像拼接运行时间三个指标对改进后的算法进行评价,并与尺度不变特征变换算法等进行对比。数据分析与拼接结果表明,使用拓扑约束去除错误匹配后特征点的重复率为98.6%,与待拼接图像中具有最多重复的特征点,从而达到了提高匹配的准确性和鲁棒性以及图像拼接精度的效果,拓扑约束后特征点辨识率达到0.83的平均水平,算法去除错误匹配对,提高了特征点的匹配正确性。该算法继承了SIFT算法强壮的稳健性,进一步提高了图像拼接的准确性和真实性。
In view of the fact that there are a lot of mismatches in the application of scale invariant feature transform (SIFT) in image stitching,we propose an improved SIFT algorithm to eliminate mismatch based on topology theory. Defining topological structure between feature points,the difference or operation of topological relation between feature points of two images is used to obtain the judgment matrix,and the set of feature points with the same topological relation is obtained,which effectively improves the matching accuracy of feature points. In addition,we evaluate the improved algorithm from three indexes:the repetition rate of feature points,the recognition rate of feature points and the running time of image stitching,and make a comparison with the scale-invariant feature transformation algorithm. The results of data analysis and stitching show that the repetition rate of the feature points removed by topology constraints is 98.6%,which is the most repetitive feature points in the image to be stitched,so that the accuracy and robustness of the matching can be improved. In order to improve the accuracy of image stitching,the recognition rate of feature points reaches an average level of 0.83 after topological constraints. The algorithm removes the error matching pairs,improves the correctness of feature points matching and inherits the robust robustness of SIFT algorithm. The accuracy and authenticity of image stitching are further improved.
作者
周露露
路纲
田艳玲
ZHOU Lu-lu;LU Gang;TIAN Yan-ling(School of Computer Science,Shaanxi Normal University,Xi'an 710000,China)
出处
《计算机技术与发展》
2019年第6期37-41,共5页
Computer Technology and Development
基金
中央高校基本科研业务费创新团队(GK201801004)
陕西省自然科学基础研究计划项目(2017JM6103,2017JM6060)
陕西师范大学2017年度校级综合教改研究项目(17JG33)
关键词
特征匹配
图像拼接
尺度不变特征变换
拓扑约束
重复率
辨识率
feature matching
image stitching
scale invariant feature transform
topological constraint
repetition rate
recognition rate
