双重降维HOG结合SVM的快速手指静脉识别

褚洪佳; 陈光化; 汪凯旋

双重降维HOG结合SVM的快速手指静脉识别

褚洪佳^1,,
陈光化^{1, 2},
汪凯旋¹

1.
上海大学微电子研究与开发中心, 上海 200444
2.
上海大学机电工程与自动化学院, 上海 200444

基金项目:

国家自然科学基金项目 61671285

详细信息

作者简介:
褚洪佳（1994-），男，山东省枣庄市人，硕士研究生，研究方向为图像处理、模式识别。E-mail: chu_hongjia@163.com

中图分类号: TP391.4
计量
- 文章访问数: 123
- HTML全文浏览量: 36
- PDF下载量: 18
- 被引次数: 0
出版历程
- 收稿日期: 2021-01-23
- 修回日期: 2021-04-08
- 刊出日期: 2022-03-20

Fast Finger Vein Recognition Based on a Dual Dimension Reduction Histogram of Oriented Gradient and Support Vector Machine

1.
Microelectronics Research & Development Center, Shanghai University, Shanghai 200444, China
2.
School of Mechatronic Engineering and Automation, Shanghai University, Shanghai 200444, China

摘要

摘要: 为减少手指静脉识别时间，提出一种双重降维方向梯度直方图特征（Histogram of Oriented Gradient，HOG）结合支持向量机（Support Vector Machine，SVM）分类的手指静脉识别方法。针对传统HOG算法特征维数高的问题，首先通过Fisher准则衡量梯度方向区间HOG特征的分类能力，然后使用序列前向选择法挑选出分类能力较优异的梯度方向区间构建部分方向区间HOG特征，最后使用主成分分析（Principal Component Analysis，PCA）降维。在公开的手指静脉数据库FV-USM和THU-FV上使用SVM多分类器进行分类识别，实验结果表明：双重降维HOG方法相较于HOG+PCA方法提取的特征维数降低了40%，识别时间减少了29.85%，识别准确率分别为99.17%和100%，等误率分别为1.07%和0.01%。
- 手指静脉识别 /
- 方向梯度直方图 /
- 特征选择 /
- 主成分分析 /
- 支持向量机
Abstract: An identification model using a dual-dimension reduction histogram of oriented gradients (HOG) combined with a support vector machine (SVM) is proposed to reduce the time required for finger vein recognition. To solve the problem of high feature dimensionality in the traditional HOG algorithm, the classification ability of the gradient direction interval is first measured using the Fisher criterion. Next, the sequence forward selection method is used to select the gradient direction interval with optimal classification ability to construct a partial direction interval HOG feature. Finally, principal component analysis (PCA) is used to reduce the number of dimensions. An SVM multi-classifier was used for the classification of the FV-USM and THU-FV datasets. The experimental results demonstrate that compared to the HOG+PCA method, the feature dimensions extracted by the dual-dimensional reduction HOG method are reduced by 40%, the recognition time is reduced by 29.85%, the recognition accuracy is 99.17% and 100%, respectively, and the equal error rate is 1.07% and 0.01%, respectively.
- finger vein recognition /
- HOG /
- feature selection /
- PCA /
- SVM

HTML全文

图 1 手指静脉识别总体结构图

Figure 1. Finger vein recognition structure

下载: 全尺寸图片幻灯片

图 2 手指静脉图像预处理过程

Figure 2. Preprocessing of finger vein image

下载: 全尺寸图片幻灯片

图 3 边缘检测Sobel模板

Figure 3. Sobel template for edge detection

下载: 全尺寸图片幻灯片

图 4 HOG算法中的sobel模板

Figure 4. Sobel template in HOG algorithm

下载: 全尺寸图片幻灯片

图 5 序列前向选择法流程图

Figure 5. Flow chart of sequential forward selection method

下载: 全尺寸图片幻灯片

图 6 惩罚因子与准确率关系图

Figure 6. Relationship between C and accuracy

下载: 全尺寸图片幻灯片

图 7 基于SVM的四分类器

Figure 7. Four classifiers based on SVM

下载: 全尺寸图片幻灯片

图 8 方向区间数目与识别准确率关系图

Figure 8. Relationship between the number of direction intervals and the recognition accuracy

下载: 全尺寸图片幻灯片

图 9 ROC曲线图（FV-USM数据库）

Figure 9. ROC Curves (FV-USM Database)

下载: 全尺寸图片幻灯片

图 10 不同数据库手指静脉图像

Figure 10. Finger vein images from different databases

下载: 全尺寸图片幻灯片

表 1 核函数分类训练数据

Table 1. Training data of kernel function

Kernel	Number of support vectors		Accuracy
Kernel	MIN	MAX	Accuracy
Linear	5	8	98.67%
Polynomial	8	8	75.17%
RBF	8	8	97.83%
Sigmoid	8	8	97.83%

下载: 导出CSV

表 2 时间开销比较

Table 2. Comparison of time cost

Methods	Dimension of feature	Extraction time/ms	Recognition time/ms
LBP+SVM	1152	798.64	98.32
HOG+SVM	672	25.73	35.73
HOG+PCA+SVM	145	26.76	19.10
Proposed method	87	26.97	13.59

下载: 导出CSV

表 3 不同方法实验数据比较

Table 3. Comparison of experimental data of different methods

Methods	FV-USM		THU-FV
	Accuracy/%	EER/%	Accuracy/%	EER/%
Maximum curvature point+MHD	87.64	10.46	98.38	2.12
Wide line detector+LTS-HD	92.38	5.35	99.26	0.34
LBP+ Euclidean distance	96.13	3.66	100	0.32
Dual dimension reduction HOG+ Euclidean distance	97.26	4.12	100	0.28
LBP+SVM	97.86	2.16	100	0.04
Proposed method	99.17	1.07	100	0.01

下载: 导出CSV

参考文献(14)

[1]	YANG L, YANG G P, YIN Y L, et al. Finger vein recognition with anatomy structure analysis[J]. IEEE Transactions on Circuits & Systems for Video Technology, 2018, 28(8): 1892-1905. https://ieeexplore.ieee.org/document/7882665
[2]	CHEN G H, DAI QH, TANG X, et al. An improved least trimmed square hausdorff distance finger vein recognition[C]//International Conference on Systems and Informatics (ICSAI), 2018: 939-943.
[3]	LI S Y, ZHANG H G, YANG JF. Finger vein recognition based on local graph structural coding and CNN[C]//Proc of SPIE, 2019, 11069: 110693I-110693I-8.
[4]	ZHANG Y K, LI W J, ZHANG L P, et al. Adaptive Gabor convolutional neural networks for finger-vein recognition[C]//International Conference on High Performance Big Data and Intelligent Systems (Hpbd & Is), 2019: 219-222.
[5]	LIU H Y, YANG L, YANG G P, et al. Discriminative binary descriptor for finger vein recognition[J]. IEEE Access, 2018, 6: 5795-5804. doi: 10.1109/ACCESS.2017.2787543
[6]	WANG X, WANG H B, HE Y, et al. Novel Algorithm for finger vein recognition based on inception-Resnet module[J]. Proc of SPIE, 2019, 11179: 111791D-111791D-9.
[7]	陶志勇, 胡亚磊, 林森. 基于改进AlexNet的手指静脉识别[J]. 激光与光电子学进展, 2020, 57(8): 58-66. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202008007.htm TAO Zhiyong, HU Yalei, LIN Sen. Finger vein recognition based on improved AlexNet[J]. Laser & Optoelectronics Progress, 2020, 57(8): 58-66. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202008007.htm
[8]	刘超, 王容川, 许晓伟, 等. 基于改进LBP的手指静脉识别算法[J]. 计算机仿真, 2019, 36(1): 381-386. doi: 10.3969/j.issn.1006-9348.2019.01.079 LIU Chao, WANG Rongchuan, XU Xiaowei, et al. Finger vein recognition algorithm based on improved LBP[J]. Computer Simulation, 2019, 36(1): 381-386. doi: 10.3969/j.issn.1006-9348.2019.01.079
[9]	李菲, 李小霞, 周颖玥. 基于改进HOG特征和稀疏表示的手指静脉识别[J]. 传感器与微系统, 2018, 37(11): 38-41. https://www.cnki.com.cn/Article/CJFDTOTAL-CGQJ201811011.htm LI Fei, LI Xiaoxia, ZHOU Yingyue. Finger vein recognition based on improved HOG features and sparse representation[J]. Transducer and Microsystem Technologies, 2018, 37(11): 38-41, 44. https://www.cnki.com.cn/Article/CJFDTOTAL-CGQJ201811011.htm
[10]	Veluchamy S, Karlmarx L R. System for multimodal biometric recognition based on finger knuckle and finger vein using feature-level fusion and k-support vector machine classifier[J]. IET Biometrics, 2017, 6(3): 232-242. doi: 10.1049/iet-bmt.2016.0112
[11]	徐子豪, 陈光化, 傅志威. 改进型LDA结合LBP的手指静脉识别[J]. 现代电子技术, 2020, 43(12): 1-4. https://www.cnki.com.cn/Article/CJFDTOTAL-XDDJ202012002.htm XU Zihao, CHEN Guanghua, FU Zhiwei. Finger vein recognition of improved LDA combined with LBP[J]. Modern Electronics Technique, 2020, 43(12): 1-4. https://www.cnki.com.cn/Article/CJFDTOTAL-XDDJ202012002.htm
[12]	徐铸业, 赵小强. 基于Agast-Adaboost的图像匹配算法[J]. 兰州理工大学学报, 2020, 46(4): 110-115. doi: 10.3969/j.issn.1673-5196.2020.06.001 XU Zhuye, ZHAO Xiaoqiang. Image matching algorithm based on Agast-Adaboost[J]. Journal of Lanzhou University of Technology, 2020, 46(6): 1-4. doi: 10.3969/j.issn.1673-5196.2020.06.001
[13]	贾楚. 基于改进HOG特征的行人检测算法研究[D]. 秦皇岛: 燕山大学, 2016. JIA Chu. Research of Pedestrian Detection Based on Improved HOG Features[D]. Qinhuangdao: Yanshan University, 2016.
[14]	蒋政, 程春玲. 基于Haar特性的改进HOG的人脸特征提取算法[J]. 计算机科学, 2017, 44(1): 303-307. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJA201701057.htm JIANG Zheng, CHENG Chunling. Improved HOG face feature extraction algorithm based on Haar characteristics[J]. Computer Science, 2017, 44(1): 303-307. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJA201701057.htm