Adaptive linear regression for single-sample face recognition

The single sample per person problem (SSPP) is quite common in real-world face recognition applications. In such circumstance, the lack of enough training samples often results in poor generalization ability for majority of the existing state-of-the-art methods. To address this problem, in this pape...

Full description

Saved in:
Bibliographic Details
Published inNeurocomputing (Amsterdam) Vol. 115; pp. 186 - 191
Main Authors Wang, Biao, Li, Weifeng, Li, Zhimin, Liao, Qingmin
Format Journal Article
LanguageEnglish
Published Amsterdam Elsevier B.V 04.09.2013
Elsevier
Subjects
Applied sciences
Artificial intelligence
Computer science; control theory; systems
Estimators
Exact sciences and technology
Face recognition
Galleries
Illumination
Least squares method
Linear regression
Pattern recognition. Digital image processing. Computational geometry
Regression
Single sample per person (SSPP)
State of the art
Training
Face recognition
Single sample per person (SSPP)
Linear regression
Computer vision
Image processing
Regression analysis
Adaptive method
Linear model
Luminance
Experimental result
Least squares method
Classification
Facies
Illumination
Online AccessGet full text

Cover

More Information
Summary:The single sample per person problem (SSPP) is quite common in real-world face recognition applications. In such circumstance, the lack of enough training samples often results in poor generalization ability for majority of the existing state-of-the-art methods. To address this problem, in this paper, a fairly simple but effective approach, called adaptive linear regression classifier (ALRC), is presented based on the simple observation that similar subjects have similar intra-personal variations. ALRC is a linear model representing a probe image as a linear combination of the single class-specific gallery and the intra-personal variations adaptively pulled from his/her kNNs in an auxiliary generic training set with multiple samples per person. ALRC can be easily employed with a regularized least square estimator and the decision is ruled in favor of the class with the minimum reconstruction error. Experimental results on AR and FERET face datasets show that ALRC outperforms several state-of-the-art approaches and demonstrates promising abilities against variations including expression, illumination and disguise.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0925-2312
1872-8286
DOI:10.1016/j.neucom.2013.02.004