Using the idea of the sparse representation to perform coarse-to-fine face recognition

  • Authors:

  • Yong Xu;Qi Zhu;Zizhu Fan;David Zhang;Jianxun Mi;Zhihui Lai

  • Affiliations:

  • Bio-Computing Research Center, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen, China and Key Laboratory of Network Oriented Intelligent Computation, Shenzhen, China;Bio-Computing Research Center, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen, China;Bio-Computing Research Center, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen, China and School of Basic Science, East China Jiaotong University, Nanchang, Jiangxi, China;Biometrics Research Centre, Department of Computing, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong;Bio-Computing Research Center, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen, China;Bio-Computing Research Center, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen, China

  • Venue:
  • Information Sciences: an International Journal
  • Year:
  • 2013

Quantified Score

Hi-index 0.07

Visualization

Abstract

In this paper, we propose a coarse-to-fine face recognition method. This method consists of two stages and works in a similar way as the well-known sparse representation method. The first stage determines a linear combination of all the training samples that is approximately equal to the test sample. This stage exploits the determined linear combination to coarsely determine candidate class labels of the test sample. The second stage again determines a weighted sum of all the training samples from the candidate classes that is approximately equal to the test sample and uses the weighted sum to perform classification. The rationale of the proposed method is as follows: the first stage identifies the classes that are ''far'' from the test sample and removes them from the set of the training samples. Then the method will assign the test sample into one of the remaining classes and the classification problem becomes a simpler one with fewer classes. The proposed method not only has a high accuracy but also can be clearly interpreted.