电子学报 ›› 2017, Vol. 45 ›› Issue (8): 1902-1910.DOI: 10.3969/j.issn.0372-2112.2018.08.014

• 学术论文 • 上一篇    下一篇

基于图像显著轮廓的目标检测

毕威, 黄伟国, 张永萍, 高冠琪, 朱忠奎   

  1. 苏州大学城市轨道交通学院, 江苏苏州 215131
  • 收稿日期:2016-03-21 修回日期:2016-08-01 出版日期:2017-08-25
    • 通讯作者:
    • 黄伟国
    • 作者简介:
    • 毕威,男,1992年生于安徽寿县,硕士研究生,研究方向为图像处理与计算机视觉.E-mail:wei_bi@126.com
    • 基金资助:
    • 国家自然科学基金 (No.51405320); 苏州自然科学基金 (No.SYG201511)

Object Detection Based On Salient Contour of Image

BI Wei, HUANG Wei-guo, ZHANG Yong-ping, GAO Guan-qi, ZHU Zhong-kui   

  1. School of Urban Rail Transportation, Soochow University, Suzhou, Jiangsu 215131, China
  • Received:2016-03-21 Revised:2016-08-01 Online:2017-08-25 Published:2017-08-25

摘要: 为了应对由复杂场景和目标形变所造成的目标难以检测的问题,提出一种基于图像显著性轮廓的目标检测方法.该方法首先利用全局概率边界算法(Globalized probability of boundary,gPb)提取图像轮廓,然后利用改进的最大类间方差法(Otsu)自适应地阈值处理获得图像的显著性轮廓;再通过检测并移除目标不稳定轮廓部分构造目标的鲁棒扇形模型;最后联合轮廓的多种局部及全局特征提出三种相似且基于特征概率密度分布的匹配策略,分别检测目标、目标镜面翻转以及发生旋转的目标.通过对多个数据库的实验分析,该方法能够有效地检测出目标及目标镜面翻转,同时在小偏转角范围有效检测旋转后的目标.

关键词: 轮廓提取, 扇形模型, 形状匹配, 目标检测

Abstract: To address detection difficulty caused by complex scene and the deformation of the object,the salient contour of image based method is proposed in this work.Firstly,the outline of image is extracted by Globalized probability of boundary,and then the improvement of the Otsu for adaptive threshold processing is used to attain the salient contour.Next,the robust fan shape model of the object is constructed by detecting and removing the unstable contour of the object.Finally,three similar matching strategies are proposed by jointing several local and global features of contour.They are based on the features probability density distribution and can detect object,its mirror flip and rotated object,respectively.Through the experimental analysis of multiple databases,the proposed method can effectively detect object and its mirror flip,and at the same time,it can also effectively detect the rotated object in the small range of rotation angle.

Key words: contour extraction, fan shape model, shape matching, object detection

中图分类号: