计算机应用 ›› 2012, Vol. 32 ›› Issue (07): 1890-1893.DOI: 10.3724/SP.J.1087.2012.01890

• 图形图像技术 • 上一篇    下一篇

非采样Contourlet变换与局部二值模式相结合的人脸识别

岳许要1,杨恢先1,祝贵1,冷爱莲2,李利3   

  1. 1. 湘潭大学 材料与光电物理学院,湖南 湘潭411105
    2. 湘潭大学 能源工程学院,湖南 湘潭 411105
    3. 湘潭大学 信息工程学院,湖南 湘潭 411105
  • 收稿日期:2011-12-05 修回日期:2012-01-19 发布日期:2012-07-05 出版日期:2012-07-01
  • 通讯作者: 岳许要
  • 作者简介:岳许要(1984-),男,河南郑州人,硕士研究生,主要研究方向:图像处理、数字信号处理、模式识别;杨恢先(1963-),男,湖南益阳人,教授,主要研究方向:图像处理、人工智能;祝贵(1986-),男,四川西昌人,硕士研究生,主要研究方向:图形图像处理;冷爱莲(1969-),女,湖南益阳人,讲师,主要研究方向:数字信号处理、智能控制;李利(1986-),女,湖南长沙人,硕士研究生,主要研究方向:图像处理、数字信号处理。
  • 基金资助:

    湖南省教育厅资助科研项目(10C1263);湘潭大学资助科研项目(11QDZ11)

Face recognition using nonsubsampled Contourlet transform and local binary pattern

YUE Xu-yao1,YANG Hui-xian1,ZHU Gui1,LENG Ai-lian2,LI Li3   

  1. 1. Faculty of Material and Photoelectronic Physics, Xiangtan University, Xiangtan Hunan 411105, China
    2. Energy Engineering College, Xiangtan University, Xiangtan Hunan 411105, China
    3. College of Information Engineering, Xiangtan University, Xiangtan Hunan 411105, China
  • Received:2011-12-05 Revised:2012-01-19 Online:2012-07-05 Published:2012-07-01
  • Contact: YUE Xu-yao

摘要: 针对人脸识别中姿态、光照和表情变化带来的识别率有限的问题,提出了一种基于非采样Contourlet变换(NSCT)与局部二值模式(LBP)的人脸特征提取方法。首先对人脸图像进行非采样Contourlet变换,得到多尺度、多方向的子带系数矩阵,然后利用LBP算子从每个子带系数矩阵上抽取局部邻域关系,得到各子带的LBP特征图谱,最后将这些图谱分块统计并级联后作为人脸的识别特征。利用多通道最近邻分类器的分类结果表明,所提方法能有效提高识别率,所提取的特征对光照、表情和姿态等变化具有更好的鲁棒性。

关键词: 人脸识别, 非采样Contourlet变换, 局部二值模式, 多通道最近邻分类器

Abstract: Concerning the problem of limited recognition rate caused by variations in position, illumination and expression in face recognition, an efficient face recognition method based on Nonsubsampled Contourlet Transform (NSCT) and Local Binary Pattern (LBP) was proposed. Firstly, a face image was decomposed with NSCT, and NSCT coefficients in different scales and various orientations were obtained; LBP operator was then used to get LBP feature maps by extracting local neighboring relationship from NSCT coefficients; Finally, feature maps were respectively divided into several blocks, the concatenated histograms, which were calculated over each block, were used as the face features. The experimental results using multi-channel nearest neighbor classifier based on Euclidean distance show that, the proposed method can improve the recognition rate effectively, and the extracted feature is robust to variations of illumination, face expression and position.

Key words: face recognition, Nonsubsampled Contourlet Transform (NCST), Local Binary Pattern (LBP), multi-channel nearest neighbor classifier

中图分类号: