作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 图形图像处理 • 上一篇    下一篇

基于自适应增强的图像二值描述子

卢来1,王军民1,范锐2   

  1. (1.广东海洋大学 寸金学院,广东 湛江 524094; 2.广东海洋大学 信息学院,广东 湛江 524088)
  • 收稿日期:2015-05-06 出版日期:2016-06-15 发布日期:2016-06-15
  • 作者简介:卢来(1975-),男,实验师、硕士研究生,主研方向为图形图像处理;王军民,副教授、硕士;范锐,教授。
  • 基金资助:

    广东省教育厅2014年特色创新基金资助项目(2014GXJK181)。

Binary Descriptor for Images Based on Adaboost

LU Lai  1,WANG Junmin  1,FAN Rui  2   

  1. (1.Cunjin College,Guangdong Ocean University,Zhanjiang,Guangdong 524094,China;2.School of Information,Guangdong Ocean University,Zhanjiang,Guangdong 524088,China)
  • Received:2015-05-06 Online:2016-06-15 Published:2016-06-15

摘要:

针对经典的尺度不变特征变换和快速鲁棒特征描述子存在空间占用和参数自适应学习能力较差的问题,提出一种基于自适应增强的图像二值描述子,采用优化学习的思路获取图像描述子。使用学习方法得到图像描述子的通用框架,在基于阈值响应的相似度函数上,给出一种改进的相似度函数,通过该函数可快速学习图像的描述子及二值描述子。运用图像的梯度特征构建弱学习器,通过自适应增强方法寻找弱学习器的最优权重和非线性特征响应,得到区分性强且鲁棒性好的局部特征描述子。图像匹配实验结果表明,该图像二值描述子占用存储空间少、匹配性能好。

关键词: 描述子, 图像描述, 自适应增强, 图像匹配, 尺度不变特征变换, 快速鲁棒特征, 局部特征, 弱学习器

Abstract:

Classic descriptors such as Scale Invariant Feature Transform(SIFT) and Speeded up Robust Feature(SURF) have some drawbacks in storage capacity and parameter adaptive learning,so a binary descriptor for images based on Adaboost is proposed,which can obtain image descriptor from optimal learning.A general framework using the learning method to obtain the image descriptor is developed,and a modified similarity function is presented on the basis of similarity function based on threshold response,by which the image descriptors and binary descriptors can be quickly learned.Weak learners are constructed by using the gradient features of the image,and the optimal weights and non-linear characteristic response of weak learners are computed by using the Adaboost method.The resulting local feature descriptor is discriminative and robust.Experimental results on image matching show that the proposed binary descriptor occupies less storage space and has good matching performance.

Key words: descriptor, image description, Adaboost, image matching, Scale Invariant Feature Transform(SIFT), Speeded up Robust Feature(SURF), local feature, weak learner

中图分类号: