图像直线段检测算法综述与测评  

Overview and Evaluation of Image Straight Line Segment Detection Algorithms

作　　者：郑行家 钟宝江[1] ZHENG Hangjia;ZHONG Baojiang(School of Computer Science and Technology, Soochow University, Suzhou, Jiangsu 215000, China)

机构地区：[1]苏州大学计算机科学与技术学院

出　　处：《计算机工程与应用》2019年第17期9-19,共11页Computer Engineering and Applications

基　　金：国家自然科学基金（No.61572341）;江苏高校优势学科建设工程资助项目

摘　　要：对现有的直线段检测算法进行系统的综述和测评。在综述方面,将现有算法划分为两类:全局Hough变换法和局部感知组合法。分析了这两类算法的技术基础和实现方案,并对其优缺点进行评述。在测评方面,Cho等人在近期的一项工作中首次建立了一个客观测评系统。但遗憾的是,该测评系统在代码实现和实际使用两个方面均存在比较严重的错误,因此所得出的结果没有能够正确反映当前各种算法的性能。全面修正了这一测评系统的错误,并对当前最具代表性的八款直线段检测算法进行评估和比较。展望直线段检测技术未来的发展趋势,为相关研究提供参考和借鉴。The existing line segment detectors are systematically reviewed and evaluated. In the overview part, the existing detectors are divided into two categories:the global Hough-transform-based method and the local perceptual grouping method. The technical basis and implementation mechanisms of these two kinds of algorithms are investigated, and their advantages and disadvantages are summarized. In the evaluation part, Cho et al. established the first objective evaluation system for line segment detectors in a recent work. Unfortunately, this evaluation system has serious bugs in both code implementation and practical application. In consequence, the results obtained in the previous work of Cho et al. have not given an accurate enough picture of the detection performance of the current algorithms. The bugs in the evaluation system are all corrected, and 8 state-of-the-art line segment detectors are evaluated and compared. Finally, the prospects of future trends in this field are discussed in order for facilitating the relevant research.

关 键 词：直线段检测 HOUGH变换 感知组合 客观测评 主观测评 精度 召回率 F-值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]