计算机工程 ›› 2018, Vol. 44 ›› Issue (7): 259-263,270.doi: 10.19678/j.issn.1000-3428.0047613

• 图形图像处理 • 上一篇    下一篇

基于Zernike矩与BoF-SURF特征融合的花粉图像分类识别

谢永华 a,b,朱延刚 a,赵贤国 b   

  1. 南京信息工程大学 a.计算机与软件学院; b.江苏省网络监控中心,南京 210044
  • 收稿日期:2017-06-15 出版日期:2018-07-15 发布日期:2018-07-15
  • 作者简介:谢永华(1976—),男,教授、博士,主研方向为基于内容的图像检索、模式识别;朱延刚、赵贤国,硕士研究生。
  • 基金项目:

    国家自然科学基金(61375030)。

Classification and Identification of Pollen Images Based on Zernike Moment and BoF-SURF Feature Fusion

XIE Yonghua a,b,ZHU Yangang a,ZHAO Xianguo b   

  1. a.School of Computer and Software; b.Jiangsu Engineering Center of Network Monitoring, Nanjing University of Information Science and Technology,Nanjing 210044,China
  • Received:2017-06-15 Online:2018-07-15 Published:2018-07-15

摘要:

针对传统单一花粉图像鉴别特征普遍存在抗噪声干扰能力弱、几何不变性低等问题,提出一种融合Zernike矩全局特征和加速鲁棒性特征包BoF-SURF局部斑点特征的花粉图像分类识别算法。提取花粉图像的Zernike矩描述子以及基于尺度空间梯度信息的SURF特征描述子,使 用K-means聚类算法对SURF特征描述子进行特征聚类,构建花粉图像的SURF视觉特征包,并对2种特征进行融合用于花粉图像分类识别。实验结果表明,与传统的花粉图像特征提取算法相比,该算法对花粉尺度和旋转变化具有较好的鲁棒性,在Confocal和Pollenmonitor图像数 据集上均获得了较高的识别率。

关键词: Zernike矩, 目标识别, K-means聚类算法, 特征融合, 花粉识别

Abstract:

Aiming at the problem of weak noise interference ability and low geometric invariance of traditional single pollen discriminative feature,this paper proposed anew pollen image classification algorithm combining global Zernike moments and bag of features from speeded up robust features BoF-SURF local speckle feature.The Zernike moment descriptor of the pollen image is extracted,the SURF feature descriptor based on the scale spatial gradient information is extracted,which is further clustered by the K-means clustering algorithm for constructing the SURF visual features bag(BoF,Bag of Visual Features)of the pollen image.The two features are integrated for pollen images classification.Experimental results show that compared with traditional pollen image feature extraction algorithm,the integrated features have good robustness to the scale and rotation of the pollen,and a high recognition rate is obtained on the Confocal and Pollenmonitor image datasets.

Key words: Zernike moment, object identification, K-means clustering algorithm, feature fusion, pollen identification

中图分类号: