一种基于阶阵列的BRIEF特征描述子  

BRIEF Feature Descriptor Based on Order Array

作　　者：张娓娓[1] 赵金龙[1] 何佳 陈绥阳[1,2] 王杰 ZHANG Wei-wei;ZHAO Jin-long;HE Jia;CHEN Sui-yang;WANG Jie(School of Electronics and Information Engineering,Xi’an Siyuan University,Xi’an 710038,China;School of Science,Xi’an Jiaotong University,Xi’an 710071,China)

机构地区：[1]西安思源学院电子信息工程学院,陕西西安710038 [2]西安交通大学理学院,陕西西安710071

出　　处：《计算机技术与发展》2023年第5期81-87,共7页Computer Technology and Development

基　　金：陕西省教育厅自然科学研究基金资助项目(18JK1104);陕西省教育科学规划项目(SGH21Y0322)。

摘　　要：局部特征匹配是机器视觉研究领域中的一个基础问题,也是该领域的研究热点之一,在目标识别、目标跟踪、场景区分等应用中具有重要的作用。而在局部特征匹配研究过程中,如何在满足多种图像变换的前提下,设计一种高效的图像特征描述子是需要解决的一个关键问题。现有的特征描述子,如SIFT和SURF,计算复杂性较高,难以胜任实时视频或移动计算环境;BRIEF特征描述子计算简单,匹配效率高,能满足实时视频或者移动计算环境的要求,但其仅考虑了单个像素,不具备方向,也就不具有旋转不变性。在BRIEF特征描述子的基础上,该文选择多个特征点,并引入阶排列方法,提出一种改进的特征描述子OPoBRIEF。相对于传统的特征描述子,OPoBRIEF能够包含更多的局部特征信息,并且计算复杂性较低。通过特征描述子稳定性实验,表明OPoBRIEF比BRRIEF具有更高的匹配正确率和更好的稳定性。而特征描述子旋转不变性的实验则表明,在旋转角度为10~12区间,OPoBRIEF与SIFT效果相当,但明显优于ORB算法。Local feature matching is a basic and hot problem in the area of computer vision,which plays an important role in many application fields such as object recognition,visual tracking,scene classification and so on.The key to this problem is how to design an effective image feature descriptor regarding for different image deformations in the process of local feature matching research.Existing feature descriptors,such as SIFT and SURF,lack the ability for real-time or mobile applications due to their high complexity.BRIEF feature descriptor is simple in calculation and has high matching efficiency,which can meet the requirements of real-time video or mobile computing environment.However,it only considers a single pixel and does not have direction,so it does not have rotation invariance.Based on the BRIEF,we select several feature points and introduce the order arrangement method to propose an improved feature descriptor OPoBRIEF.Compared with traditional feature descriptors,OPoBRIEF can contain more local feature information and has lower computational complexity.The feature descriptor stability experiments show that OPoBRIEF has higher matching accuracy and better stability than BRRIEF.The experiment on the rotation invariance of feature descriptors shows that OPoBRIEF has the same effect as SIFT in the rotation angle range of 10~12,but it is obviously better than ORB algorithm.

关 键 词：描述子 二值模式 BRIEF 阶排列 旋转不变性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]