华南理工大学学报(自然科学版) ›› 2013, Vol. 41 ›› Issue (11): 8-15,22.doi: 10.3969/j.issn.1000-565X.2013.11.002

• 电子、通信与自动控制 • 上一篇    下一篇

多向格型微分和竞争抑制的角点检测改进方法

马丽红1 黎剑晖1 谭幸均2   

  1. 1.华南理工大学 电子与信息学院,广东 广州 510640;2.海思半导体有限公司 网络架构与系统设计部,广东 深圳 518129
  • 收稿日期:2013-05-21 修回日期:2013-07-11 出版日期:2013-11-25 发布日期:2013-10-11
  • 通信作者: 马丽红(1965-),女,博士,教授,主要从事图像视频信号处理、容错编码和数据隐藏、模式识别研究. E-mail:eelhma@scut.edu.cn
  • 作者简介:马丽红(1965-),女,博士,教授,主要从事图像视频信号处理、容错编码和数据隐藏、模式识别研究.
  • 基金资助:

    国家自然科学基金资助项目(60972133)

Improved Corner Detection Based on Multi- Direction Latticed Differential and Competitive Suppression

Ma Li- hong1 Li Jian- hui1 Tan Xing- jun2   

  1. 1.School of Electronic and Information Engineering,South China University of Technology,Guangzhou 510640,Guangdong,China;2.Department of Network Architecture and System Design,HiSilicon Technologies Co.,Ltd.,Shenzhen 518129,Guangdong,China
  • Received:2013-05-21 Revised:2013-07-11 Online:2013-11-25 Published:2013-10-11
  • Contact: 马丽红(1965-),女,博士,教授,主要从事图像视频信号处理、容错编码和数据隐藏、模式识别研究. E-mail:eelhma@scut.edu.cn
  • About author:马丽红(1965-),女,博士,教授,主要从事图像视频信号处理、容错编码和数据隐藏、模式识别研究.
  • Supported by:

    国家自然科学基金资助项目(60972133)

摘要: 为了去除图像锯齿边缘的伪角点,减少平滑缓变角点的重复检出,提出了多向格型微分( DLD) 算子,结合“胜者全取”( WTA) 的竞争抑制策略,改善角点检测的性能.首先,针对锯齿失真,提出一种基于Directionlet 分解的微分算子,其多向格型复合表示能实现任意方向图像边缘的无锯齿结构分解,很好地抑制锯齿特征的伪角点现象.其次,针对缓变角点,提出竞争响应抑制法,它根据DLD 响应值定义一个以邻域尺寸为约束条件的角点相似度,相似角点通过WTA 策略合并。实验结果表明,联合DLD 的GLCP 检测器和FAST 检测器能分别使误检和漏检综合评价指标ACU 值提高12.14% 和6.32%,改善效果明显.

关键词: 角点检测, 格型微分, 竞争抑制, Directionlet 分解, 伪角点, 胜者全取

Abstract:

A multi- direction latticed differential (DLD) operator combined with the“Winner Takes All (WTA)”competitive suppression strategy is proposed to remove the false corners along zigzag image edges and the repeatedly-detected corners in smooth regions,which improves the performance of the corner detector.In the investigation,first,a differential operator based on the Directionlet decomposition is proposed to eliminate the zigzag distortion.With the multi- direction latticed complex representation of the operator,image edges in arbitrary direction can bedenoted without zigzag,and false corners with zigzag appearance can be effectively suppressed.Then,a competitivesuppression algorithm,which defines a corner similarity with the constraint of neighborhood size according to theDLD response value,and merges similar corners with the WTA strategy,is put forward to remove slowly- varyingcorners.Experimental results show that the typical GLCP detector and the FAST detector combined with DLD areboth obviously improved.For instance,the ACU value,which is a comprehensive evaluation index of the error de-tection rate and the missed detection rate,increases by 12.14% and 6.32%,respectively.

Key words: corner detection, latticed differential, competitive suppression, Directionlet decomposition, false corner, Winner Takes All

中图分类号: