作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

一种应用于人脸识别的超分辨率方法

夏苏娜,马小虎   

  1. (苏州大学计算机科学与技术学院,江苏 苏州 215006)
  • 收稿日期:2013-03-13 出版日期:2014-04-15 发布日期:2014-04-14
  • 作者简介:夏苏娜(1988-),女,硕士研究生,主研方向:模式识别,图像处理;马小虎,教授。

A Super-resolution Method Applied to Face Recognition

XIA Su-na, MA Xiao-hu   

  1. (School of Computer Science and Technology, Soochow University, Suzhou 215006, China)
  • Received:2013-03-13 Online:2014-04-15 Published:2014-04-14

摘要: 为提高局部保持投影(LPP)在人脸图像超分辨率中的适用性,在LPP中引入典型相关分析(CCA),提出一种相关性增强的局部保持投影方法(CELPP)。CELPP用于提取高分辨率图像与低分辨率图像特征,根据关系学习建立低分辨率图像特征与高分辨率图像特征之间的映射变换,输入低分辨率图像,通过CELPP特征提取和关系映射,得到高分辨率图像,并将其用于人脸识别。对人脸库ORL和Yale进行的实验结果表明,该方法同时考虑了高分辨率图像与低分辨率图像的相似性及同类图像的局部结构性,在基于人脸识别的超分辨率应用中优于LPP和CCA。

关键词: 超分辨率, 局部保持投影, 典型相关分析, 相关性增强的局部保持投影, 关系学习

Abstract: To enhance the applicability of Locality Preserving Projections(LPP) in super-resolution of face images, this paper proposes an improved method, Correlation Enhanced Locality Preserving Projection(CELPP), which introduces the method of Canonical Correlation Analyses(CCA) into LPP. CELPP is used for feature extraction, and relationship learning is used to build a bridge for the transformation of high resolution images and low resolution images. Entering low resolution images, through CELPP feature extraction and mapping transformation, the high resolution images are achieved and used for face recognition. Experimental results of ORL and Yale databases show that CELPP is better than LPP and CCA in super-resolution applications because CELPP considers the similarity of high resolution images and low resolution images, and the local structure of the same class images.

Key words: super-resolution, Locality Preserving Projections(LPP), Canonical Correlation Analysis(CCA), Correlation Enhanced Locality Preserving Projections(CELPP), relationship learning

中图分类号: