Loading…
Constrained L1-optimal sparse representation technique for face recognition
•L1-CSRT is proposed for one sample per person and based on sparse representation.•Nearest neighbours are identified using kernel-based distance.•It utilizes constrained L1-Cuckoo Search Algorithm for coefficient (λ) evaluation.•Optimizing λ increases sparsity and provides higher discrimination abil...
Saved in:
Published in: | Optics and laser technology 2020-09, Vol.129, p.1, Article 106232 |
---|---|
Main Authors: | , , |
Format: | Article |
Language: | English |
Subjects: | |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | •L1-CSRT is proposed for one sample per person and based on sparse representation.•Nearest neighbours are identified using kernel-based distance.•It utilizes constrained L1-Cuckoo Search Algorithm for coefficient (λ) evaluation.•Optimizing λ increases sparsity and provides higher discrimination ability.•It is analysed with an improvement of 2–6 % in terms of classification accuracy.
The performance of face recognition (FR) system essentially depends on the availability of training data. With limited training data available, it creates a major challenge in real world applications. This motivates researchers to investigate for such technique which generates optimum results using limited training samples or even single sample per person (SSPP). In this paper, a novel FR method is developed as Constrained L1-Optimal Sparse Representation Technique (L1-CSRT) for SSPP. An optimal sparse representation technique is formulated using the constrained Cuckoo search algorithm (CSA) for estimation of λ coefficients. Further, a novel fitness function is developed based on the L1-norm for better classification accuracy yielding better FR. The motivation behind the optimization of λ coefficients using CSA is to increase the sparsity by a better exploration of search space. The efficiency and accuracy of proposed L1-CSRT is shown based on the extensive simulations carried out on standard databases. The experimental results are compared with existing methods in terms of mean classification error. The performance of L1-CSRT is analysed with an improvement of 2--6% in terms of classification accuracy. |
---|---|
ISSN: | 0030-3992 1879-2545 |
DOI: | 10.1016/j.optlastec.2020.106232 |