摘要
针对传统身份鉴别要求用户参与、需使用专用设备等弊端,提出了基于人脸图像的身份鉴别,鉴别流程依次为:图像预处理、人脸检测、人脸特征提取、特征重构和匹配五个部分。采用傅里叶变换提取全局特征,Gabor小波变换提取局部特征。提出了特征重构,并用纹理分析方法(Texture Analysis Method,TAM)重构年青纹理特征。在人脸数据库SFD中用自建的向量最大似然(Vector Maximum Likelihood,VML)分类器进行了实验,取得了95%的平均准确率。
In this paper,apersonal identification system based on facial image is presented,concerning the defects of user participation and using special equipment in traditional identification.It is composed of image preprocessing,face detection,facial feature extraction,features reconstruction and matching.Global features are extracted by discrete Fourier transform and local features are extracted by Gabor wavelets transform.Features reconstruction is presented and young texture features are reconstructed using texture analysis methods.The experiments is conducted with own Vector Maximum Likelihood classifier in the SFD facial database,an average accuracy rate of 95%is achieved.
出处
《计算机与数字工程》
2015年第10期1875-1879,共5页
Computer & Digital Engineering
基金
山西省高等学校教学改革项目(编号:J2012073)
山西大同大学科学研究项目(编号:2011k8)
山西省软科学研究计划项目(编号:2014041049-1)
山西省大同市政府专项研究项目(编号:2014001)资助
关键词
身份鉴别
纹理分析
特征重构
向量最大似然分类器
personal identification
texture analysis
features reconstruction
vector maximum likelihood classifier
