南京大学学报(自然科学版) ›› 2020, Vol. 56 ›› Issue (1): 51–56.doi: 10.13232/j.cnki.jnju.2020.01.006

• • 上一篇    下一篇

AdaBoost图像到类距离学习的图像分类方法

李子龙1,2,3(),周勇2,鲍蓉1   

  1. 1. 徐州工程学院信电工程学院,徐州,221018
    2. 中国矿业大学计算机科学与技术学院,徐州,221116
    3. 江苏昂内斯电力科技股份有限公司博士后科研工作站,徐州,221003
  • 收稿日期:2019-08-25 出版日期:2020-01-30 发布日期:2020-01-10
  • 通讯作者: 李子龙 E-mail:55001787@qq.com
  • 基金资助:
    国家自然科学基金(60273064);江苏省建设系统科技项目(2018ZD077);江苏省智慧工业控制技术重点建设实验室开放基金(JSKLII201802)

AdaBoost image⁃to⁃class distance learning for image classification

Zilong Li1,2,3(),Yong Zhou2,Rong Bao1   

  1. 1. School of Information and Electrical Engineering,Xuzhou Institute of Technology,Xuzhou,221018,China
    2. School of Computer Science and Technology,China University of Mining and Technology,Xuzhou,221116,China
    3. Post Doctoral Research Center,Onnes Power Technology Co. , Ltd,Xuzhou,221003,China
  • Received:2019-08-25 Online:2020-01-30 Published:2020-01-10
  • Contact: Zilong Li E-mail:55001787@qq.com

摘要:

近年来,距离度量学习已经成为图像分类领域的研究热点之一,图像到类距离的度量作为其中的一种方法,取得了不错的分类效果.该方法是一种非参数方法,但由于缺少训练学习,其分类性能很容易受干扰因素的影响,为此提出一种基于AdaBoost算法的图像到类距离学习的图像分类方法.首先将图像到类的距离进行阈值化处理,并使用线性分段函数作为图像到类距离的评价函数,然后将该评价函数作为弱分类器加入到AdaBoost算法中生成一个强分类器.为了选择最优的弱分类器,使用粒子群优化算法确定图像的相似性阈值,再基于权重错误误差最小化原则得到距离评价函数的两个评价值.最后通过实验验证,该方法在Scene?15和Caltech?101图像数据集上比其他方法有更好的分类效果.

关键词: 图像分类, 图像到类距离, AdaBoost, 粒子群优化算法

Abstract:

Recently,distance metric learning has become one of the most attractive research areas in image classification. The image?to?class distance metric is a non?parametric method for image classification and achieves a impressive result. However,due to the lack of a training phase,the classification accuracy of it is easily affected by irrelevant factors.In this paper,we propose a novel image?to?class distance learning method for image classification by using the AdaBoost algorithm. We first deal with image?to?class distance through the threshold,and a piecewise linear discriminator function is used as the evaluation function of image?to?class distance. Then,the evaluation function is added to the AdaBoost algorithm as a weak classifier to generate a strong classifier.In order to select the optimal weak classifier,the particle swarm optimization algorithm is used to determine the similarity threshold of the image,and the two evaluation values of the distance evaluation function are obtained based on the principle of weight error minimization.The experimental results on datasets of Scene?15 and Caltech?101 verified that our proposed method can significantly outperform other methods in image classification.

Key words: image classification, image?to?class distance, AdaBoost, particle swarm optimization algorithm

中图分类号: 

  • TP301.4

图1

图像到类距离的分类过程"

图2

Scene?15图像集中AdaBoost 算法迭代次数与分类准确率的关系"

图3

Caltech?101图像集中AdaBoost 算法迭代次数与分类准确率的关系"

图4

Scene?15图像集中每种类别上不同方法的图像分类性能对比"

图5

Caltech?101图像集中每种类别上不同方法的图像分类性能对比"

图6

不同方法在Scene?15数据集上的ROC曲线"

表1

不同方法的平均分类准确性(%)对比"

方法Scene?15Caltech?101
NBNN78.7±0.7471.6±1.12
LI2C81.7±0.5477.8±0.98
Local NBNN83.3±0.5979.4±1.10
LDC84.4±0.5280.3±0.95
SI2C86.5±0.5081.2±0.81
本文方法89.8±0.4484.7±0.78
1 Zhang L,Zhen X T,Shao L. Learning object?to?class kernels for scene classification. IEEE Transactions on Image Processing,2014,23(8):3241-3253.
2 Moutafis P,Leng M J,Kakadiaris I A. An overview and empirical comparison of distance metric learning methods. IEEE Transactions on Cybernetics,2016,47(3):612-625.
3 Rasche C. Rapid contour detection for image classification. IET Image Processing,2017,12(4):532-538.
4 Tang P,Wang X G,Feng B,et al. Learning multi?instance deep discriminative patterns for image classification. IEEE Transactions on Image Processing,2017,26(7):3385-3396.
5 Lu Y W,Lai Z H,Li X L,et al. Learning parts?based and global representation for image classification. IEEE Transactions on Circuits and Systems for Video Technology,2018,28(12):3345-3360.
6 Boiman O,Shechtman E,Irani M. In defense of nearest?neighbor based image classification∥IEEE Conference on Computer Vision and Pattern Recognition. Anchorage,AL,USA:IEEE,2008:1-8.
7 Wang Z X,Hu Y Q,Chia L T. Image?to?class distance metric learning for image classification∥European Conference on Computer Vision. Springer Berlin Heidelberg,2010:709-719.
8 Wang Z X,Hu Y Q,Chia L T. Improved learning of I2C distance and accelerating the neighborhood search for image classification. Pattern Recognition,2011,44(10-11):2384-2394.
9 McCann S,Lowe D G. Local naive bayes nearest neighbor for image classification∥2012 IEEE Conference on Computer Vision and Pattern Recognition. Providence,RI,USA:IEEE,2012:3650-3656.
10 Wang Z L,Feng J S,Yan S C,et al. Linear distance coding for image classification. IEEE trAnsactions on Image Processing,2013,22(2):537-548.
11 Wang Z L,Feng J S,Yan S C. Collaborative linear coding for robust image classification. International Journal of Computer Vision,2015,114(2-3):322-333.
12 Qiao S Y,Li Z L. Discriminative learning of I2C distance for image classification∥The 27th Chinese Control and Decision Conference. Qingdao,China:IEEE,2015:1371-1375.
13 Tan S B,Liu L,Peng C Y,et al. Image?to?class distance ratio:a feature filtering metric for image classification. Neurocomputing,2015,165:211-221.
14 Peng P. Context?aware image?to?class distances from image classification. Ph.D Dissertation. Sheffield:University of Sheffield,2017.
15 Wang H J,Hu J N,Deng W H. Face feature extraction:a complete review. IEEE Access,2017,6:6001-6039.
16 Fernández?Baldera A,Buenaposada J M,Baumela L. BAdaCost:multi?class boosting with costs. Pattern Recognition,2018,79:467-479.
[1] 洪思思,曹辰捷,王 喆*,李冬冬. 基于矩阵的AdaBoost多视角学习[J]. 南京大学学报(自然科学版), 2018, 54(6): 1152-1160.
[2] 吴家豪1,彭志平2*,崔得龙2,李启锐2,何杰光2. 基于多Agent系统的粒子群遗传优化云工作流调度算法[J]. 南京大学学报(自然科学版), 2017, 53(6): 1114-.
[3]  李 蓓,张兴敢,方 晖*.  一种改进的BP-Adaboost算法及在雷达多目标分类上的应用[J]. 南京大学学报(自然科学版), 2017, 53(5): 984-.
[4]  曹冬寅,王 琼*,张兴敢. 基于稀疏重构残差和随机森林的集成分类算法
[J]. 南京大学学报(自然科学版), 2016, 52(6): 1127-.
[5] 周文猛1,杨一品1,周余1,于耀1,金苏文2,都思丹1*. 基于Kinect的无标记手部姿态估计系统[J]. 南京大学学报(自然科学版), 2015, 51(2): 297-.
[6] 邹晓川,叶翰嘉,詹德川*. 基于强模态辅助的弱模态图像概念分类及检测[J]. 南京大学学报(自然科学版), 2014, 50(2): 228-.
[7]   韩飞**,杨春生,刘清 .  一种改进的基于梯度搜索的粒子群优化算法*[J]. 南京大学学报(自然科学版), 2013, 49(2): 196-201.
[8]  刘杨1,田学锋2,詹志辉1**
.  粒子群优化算法惯量权重控制方法的研究*
[J]. 南京大学学报(自然科学版), 2011, 47(4): 364-371.
[9]  严超 1 , 王元庆 1 ** , 张兆扬 2.  基于 AdaBoost 和 Kalman 算法的人眼检测与跟踪* 

[J]. 南京大学学报(自然科学版), 2010, 46(6): 681-687.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
[1] 王广辉,杨高东,周 政,张志炳. 丙酸异龙脑酯的合成、反应热力学和反应动力学研究[J]. 南京大学学报(自然科学版), 2019, 55(3): 486 -497 .
[2] 张龙波, 李智远, 杨习贝, 王怡博. 决策代价约简求解中的交叉验证策略[J]. 南京大学学报(自然科学版), 2019, 55(4): 601 -608 .
[3] 曹欣怡,李鹤,王蔚. 基于语料库的语音情感识别的性别差异研究[J]. 南京大学学报(自然科学版), 2019, 55(5): 758 -764 .
[4] 徐媛媛,张恒汝,闵帆,黄雨婷. 三支交互推荐[J]. 南京大学学报(自然科学版), 2019, 55(6): 973 -983 .
[5] 刘显, 陈强路, 王小林, 丘靥, 杨源显. 方解石晶体定向性对水的拉曼光谱影响的实验评估[J]. 南京大学学报(自然科学版), 2020, 56(3): 297 -307 .
[6] 杜彦男,吴孔友,刘寅,林红梅,党思思,李彦颖,徐进军. 断陷盆地边界断裂结构特征及物性差异定量评价——以车镇凹陷埕南断裂为例[J]. 南京大学学报(自然科学版), 2020, 56(3): 405 -417 .