PDF全文下载：2015050581
于海霞1,2， 齐国清1*， 付才魁1
(1.大连海事大学 信息科学技术学院，辽宁 大连 116023;2.大连理工大学城市学院，辽宁 大连 116024）
摘要： 深度图像具有能够表征物体表面三维立体信息的优势，而直线则携带了图像中重要的结构信息，对图像分割、图像识别等后续图像分析具有十分重要的意义。提出一种新的深度图像直线检测方法，由直线组成的直线框架方法，能够精确地、综合地提取深度图像的直线特征。首先，在指定邻域内对每个深度像素估计表面法向量；通过对表面法向量的相关计算得到每个像素点的边缘强度；之后，通过全局匹配措施，连点成线；最后，采用Hough变换在深度图像上标记直线。通过仿真结果对比，不难发现相较于其他直线特征提取方法，本文提出的新方法能够提供更平滑干净的直线检测结果；此外，在折痕边缘处获得单条直线，阶跃边缘处获得两条平行直线，因而可作为深度图像的3D结构信息。
关键词： 深度图像； 直线特征； 边缘检测
中图分类号： TP 391.4 文献标志码： A
Linear Feature Extraction in Range Image
YU Haixia1,2， QI Guoqing1， FU Caikui1
(1. School of Information Science and Technology, Dalian Maritime University, Dalian 116023, China;
2. City Institute, Dalian University of Technology, Dalian 116024, China)
Abstract: Compared with gray image, range image can show the surface information of 3D object. Lines, with important structural information in image, are essential to image analysis, such as image segmentation and image recognition. This paper proposes a new line detection method in range image. Lineframe which is composed of lines can precisely and comprehensively extract line feature, and fitness measure is used to evaluate the lineframe which is composed of lines. First and foremost, we estimate surface normal of every range pixel in specific neighborhood. Then, the surface normal is calculated to represent edge strength. Secondly, we connect points to be lines in the way of fitness measure. Finally, lines are marked by Hough transform in range image. Compared with other line feature extraction methods, the proposed method can provide line detection result which is smoother and cleaner. In addition, a single line is gained in the place of crease edge, while two parallel lines are gained in the place of jump edge, which can be 3D structural information of range image.
Key words: range image; feature extraction; edge detection
收稿日期： 20141101
基金项目： 中央高校基本科研业务费专项资金资助项目（3132013054）.
作者简介： 于海霞(1975—)，女，副教授，博士研究生. *通信联系人.
文章编号：16726987（2015）05058107； DOI： 10.16351/j.16726987.2015.05.021
