Enlarge the Training Set Based on Inter-Class Relationship for Face Recognition from One Image per Person

Qin Li,Jin Xue Li,Hua Jing Wang,Jane You,Zhao Ming Li

doi:10.1371/journal.pone.0068539

Abstract

In some large-scale face recognition task, such as driver license identification and law enforcement, the training set only contains one image per person. This situation is referred to as one sample problem. Because many face recognition techniques implicitly assume that several (at least two) images per person are available for training, they cannot deal with the one sample problem. This paper investigates principal component analysis (PCA), Fisher linear discriminant analysis (LDA), and locality preserving projections (LPP) and shows why they cannot perform well in one sample problem. After that, this paper presents four reasons that make one sample problem itself difficult: the small sample size problem; the lack of representative samples; the underestimated intra-class variation; and the overestimated inter-class variation. Based on the analysis, this paper proposes to enlarge the training set based on the inter-class relationship. This paper also extends LDA and LPP to extract features from the enlarged training set. The experimental results show the effectiveness of the proposed method.

Highlights

Face recognition has attracted much attention in the last two decades
We study the principals of principal component analysis (PCA), linear discriminant analysis (LDA), and locality preserving projections (LPP) and show why they cannot perform well or applicable to one sample problem
We present our analysis from the second viewpoint: why is one sample problem itself difficult? For the first time, we ascribe the difficulty of one sample problem to four reasons: 1. the training set is small; 2. one sample is not representative; 3. the intra-class variation is unknown or underestimated; and 4. the inter-class variation is overestimated

Summary

Introduction

Face recognition has attracted much attention in the last two decades. it is still an unsolved problem that needs further investigation. Many popular subspace-based feature extraction methods [2,3,4,5,6] and classifiers [7,8,9,10,11] either cannot achieve high classification accuracy, or fail to work in one sample problem. PCA, LDA, and LPP are three popular methods proposed for feature extraction in the task of face recognition These three methods and their extensions are developed based on an implicit assumption that several images (at least two) from each individual are available in the training stage. As this implicit assumption does not hold in the one sample problem, these methods cannot achieve high classification accuracy.

Methods

Findings

Discussion

Conclusion