摘要
以光学足迹图像为研究对象,采集并构建了一个包含134人在自然行走下的2680枚足迹数据集,提出一种足迹识别算法.从光学足迹图像中分别提取频域下的局部相位量化(LPQ)纹理特征和空域下的全局形态特征,并通过特征融合和优化获取相对稳定且具有区分性的足迹特征.为了提升分类器对足迹特征的辨识性能,通过度量学习的方法将足迹特征投影到新的特征空间,使同类足迹特征分布更紧凑,异类特征间分布更离散.利用马氏距离和对数函数构造度量学习核函数,结合支持向量机(SVM)分类器实现光学足迹的识别.利用134人自然行走下光学足迹数据集进行实验,通过与不同特征、不同核的SVM分类器进行对比,结果表明:本研究算法提高了足迹的识别准确率,最高达到了96.66%,能够为足迹的应用和研究提供有效参考.
The optical footprint image was taken as the research object,a dataset containing 2 680 footprints of 134 people in natural walking was collected and constructed,and a footprint identification algorithm was proposed.The local phase quantization(LPQ) texture features in the frequency domain and global morphological features in the spatial domain were extracted respectively from the optical footprint image,and the stable and distinguishable footprint features were obtained by feature fusion and optimization.In order to improve the identification performance of footprint image features in the classifier,the footprint features were projected to a new feature space by using the metric learning method,which could make the distribution of similar footprint features more compact and the distribution of heterogeneous features more discrete.The metric learning kernel function was constructed using the Markov distance and logarithmic function,and the support vector machine(SVM) classifier was used to realize the recognition of the optical footprint.Experiments were performed on the optical footprint dataset of 134 people in natural walking.By comparison with SVM classifiers of different features and different kernels,results show that the proposed algorithm improves the recognition accuracy of footprint,and the highest recognition rate reaches to 96.66%,which could provide effective references for the application and study of footprints.
作者
鲍文霞
王云飞
王年
唐俊
BAO Wenxia;WANG Yunfei;WANG Nian;TANG Jun(Key Laboratory of Intelligent Computing of Signal Processing of Ministry of Education,Anhui University,Hefei 230601,China)
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2020年第11期11-16,共6页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家重点研发计划资助项目(2018YFC0807302)
国家自然科学基金资助项目(61772032,61672032)。
关键词
足迹识别
局部相位量化特征
全局形态特征
特征融合
度量学习核函数
footprint identification
local phase quantization feature
global morphological feature
features fusion
metric learning kernel function
