测绘学报 ›› 2017, Vol. 46 ›› Issue (2): 228-236.doi: 10.11947/j.AGCS.2017.20160250

• 摄影测量学与遥感 • 上一篇    下一篇

主成分分析的匹配点对提纯方法

董杨, 范大昭, 纪松, 雷蓉   

  1. 信息工程大学, 河南 郑州 450000
  • 收稿日期:2016-05-24 修回日期:2017-01-04 出版日期:2017-02-20 发布日期:2017-03-07
  • 作者简介:董杨(1992-),男,硕士生,研究方向为数字摄影测量。E-mail:wenku34@163.com
  • 基金资助:
    国家自然科学基金(41401534);地理信息工程国家重点实验室开放基金(SKLGIE2013-M-3-1)

The Purification Method of Matching Points Based on Principal Component Analysis

DONG Yang, FAN Dazhao, JI Song, LEI Rong   

  1. Information Engineering University, Zhengzhou 450000, China
  • Received:2016-05-24 Revised:2017-01-04 Online:2017-02-20 Published:2017-03-07
  • Contact: 范大昭 E-mail:fdzcehui@163.com
  • Supported by:
    The National Natural Science Foundation of China (No.41401534),State Key Laboratory of Geographic Information Engineering (No. SKLGIE2013-M-3-1)

摘要: 传统的匹配点对提纯算法通常需要寻找小部分点集作为初始输入,再迭代求解出能够满足大多数点对约束要求的最优解,其提纯结果易陷入局部极值,造成正确匹配点对的遗漏。针对这一问题,本文引入了主成分分析思想,将整体点集作为初始输入,逐步剔除误匹配点对,稳健求解,得到更为准确的全局最优解,降低正确匹配点对的遗漏率,达到较好的提纯效果。试验表明,本文方法在一定的原始误匹配率下,能够得到整体最优解,在剔除误匹配点对的同时,能够避免或减少正确匹配点对的遗漏。

关键词: 影像匹配, 主成分分析, 奇异值分解, 提纯, RANSAC

Abstract: The traditional purification method of matching points usually uses a small number of the points as initial input. Though it can meet most of the requirements of point constraints, the iterative purification solution is easy to fall into local extreme, which results in the missing of correct matching points. To solve this problem, we introduce the principal component analysis method to use the whole point set as initial input. And thorough mismatching points step eliminating and robust solving, more accurate global optimal solution, which intends to reduce the omission rate of correct matching points and thus reaches better purification effect, can be obtained. Experimental results show that this method can obtain the global optimal solution under a certain original false matching rate, and can decrease or avoid the omission of correct matching points.

Key words: image matching, principal components analysis, singular value decomposition, purification, random sample consensus

中图分类号: