电子学报 ›› 2016, Vol. 44 ›› Issue (1): 135-142.DOI: 10.3969/j.issn.0372-2112.2016.01.020

• 学术论文 • 上一篇    下一篇

基于局部尺度特征描述和改进DTW技术的局部轮廓匹配算法

徐贵力, 赵妍, 姜斌, 王正盛, 李开宇, 郭瑞鹏   

  1. 南京航空航天大学自动化学院, 江苏南京 210016
  • 收稿日期:2014-05-26 修回日期:2014-11-13 出版日期:2016-01-25 发布日期:2016-01-25
  • 作者简介:徐贵力 男,1972年生,黑龙江佳木斯人.现为南京航空航天大学博士生导师,从事光电检测、计算机视觉方面的研究. E-mail:guilixu@163.com 赵 妍 女,1989年生,河南人新乡人.硕士研究生,从事机器视觉方面的研究. E-mail:zhao.yan109@163.com
  • 基金资助:

    国家自然科学基金(No.61473148)

Partial Contour Matching Algorithm Based on Local Scale Description and Improved DTW

XU Gui-li, ZHAO Yan, JIANG Bin, WANG Zheng-sheng, LI Kai-yu, GUO Rui-peng   

  1. Institute of Automation, Nanjing University of Aeronautics and Astronautics, Nanjing, Jiangsu 210016, China
  • Received:2014-05-26 Revised:2014-11-13 Online:2016-01-25 Published:2016-01-25

摘要:

基于轮廓的图像匹配是计算机视觉领域中的重要问题,但是目前尚未有较成熟的算法能够很好地解决局部轮廓匹配问题及非相似变换和非刚体变换引起的轮廓形变问题.根据局部轮廓结构在产生形变时具有相对稳定性的规律及融合轮廓局部信息和全局信息的轮廓描述思想,本文提出了一种具有尺度、旋转、平移不变性,形变鲁棒性和初始点无关性的局部尺度轮廓描述算法.在此基础上,针对线性匹配方法效果不佳以及传统DTW技术约束路径的线性度不满足轮廓采样特性要求的问题,提出一种基于改进DTW技术的轮廓匹配算法,即结合轮廓采样特性设置九宫格的路径约束条件,以旋转角度为参数,计算全局最佳匹配路径.实验结果表明,对于存在尺度、平移、旋转及形变关系的两轮廓,该方法能较好地实现轮廓间的局部匹配,并且其匹配准确率平均约为92%,较HD算法提高了30%,较传统DTW算法提高了26%.

关键词: 计算机视觉, 轮廓描述, 局部尺度, 局部轮廓匹配, 动态时间规整, 形变

Abstract:

Image matching based on contour is an important issue in computer vision, but now there is no mature algorithm that can solve the problems of partial contour matching and deformation caused by non-similar transformation and non-rigid transformation.According to the conclusions that partial contour structure has relative stability and a good description need to merge the contour's local and global information, a local scale description method is proposed which is scale, rotation invariant, deformation tolerant and initial point independence.On this basis, a contour matching algorithm is proposed based on improved DTW.Experimental results show that the proposed method can realize the partial matching between open contour and open contour as well as closed contour.Its matching precision is 92% on average, and improves 30% compared with HD method, improves 26% compared with traditional DTW.

Key words: machine vision, contour description, local scale, partial matching, dynamic time warping, deformation

中图分类号: