Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (20): 134-138.DOI: 10.3778/j.issn.1002-8331.1806-0330

Previous Articles     Next Articles

Vehicle-Logo Recognition Based on Ensemble Learning with Multiple LBP Features

LI Zhe, YU Mengru   

  1. School of Electronic Engineering, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
  • Online:2019-10-15 Published:2019-10-14

基于多种LBP特征集成学习的车标识别

李哲,于梦茹   

  1. 西安邮电大学 电子工程学院,西安 710121

Abstract: In view of the difficult problem of the classification of vehicle-logo images, this paper proposes a new vehicle-logo recognition method based on ensemble learning with multiple LBP features. Firstly, the location of vehicle logo is roughly positioned based on the relative position relationship between the license plate and the vehicle logo. Then, according to the background texture feature of the vehicle logo, edge detection is used by different operators to achieve background ablation, and uses the projection method to accurately determine the vehicle logo position. Finally, the vehicle logo image segmentation uses the Center Symmetric Local Binary Pattern(CSLBP) to extract the neighborhood features of each pixel, which forms the fine texture features. It uses the LBP histogram algorithm to extract the spatial structure features of vehicle logo region. SVM and BP are used to train two features separately, get the voting decision and the classification list, and obtain the output class by the weighted sum rule fused decision matrix, which constructs an optimal ensemble classifier. Experimental results show that the recognition rate of the proposed algorithm is better than that of a single feature and classifier.

Key words: vehicle-logo position, Center Symmetric Local Binary Pattern(CSLBP), Support Vector Machine(SVM), ensemble learning

摘要: 针对车标图像的分类难问题,提出基于多种LBP特征集成学习的车标识别算法。利用车牌与车标的相对位置关系粗定位车标区域;根据车标背景纹理特征使用不同的算子进行边缘检测,进而实现背景消融,采用投影方法精确确定车标位置;将车标图像分块,应用CSLBP算子提取每个像素点邻域特征,将车标所有像素点邻域特征合成精细的纹理特征,运用LBP直方图算法提取车标区域的空间结构特征,再采用SVM和BP分别训练这两种特征,得到投票决策矩阵,利用加权求和的规则融合决策矩阵,构成最优集成分类器,输出车标类别。实验结果表明,该算法的识别率明显优于单一的特征和分类器。

关键词: 车标定位, CSLBP算子, 支持向量机(SVM), 集成学习