Multiple samples dictionary learning and locality constrained coding for face recognition
Download as PDF
DOI: 10.23977/mcee2020.011
Author(s)
Ling Gan, and Ruifang Wang
Corresponding Author
Ruifang Wang
ABSTRACT
This paper proposes a new dictionary learning algorithm, Multiple Samples Dictionary Learning and Locality Constrained Coding algorithm (MSDL-LCC), to solve the problems that the insufficient number of training samples when learning a dictionary and the lack of discriminative power of the test coefficient. The proposed algorithm first generates virtual training samples for the origin training data, and then uses all the training samples to learn a dictionary. Finally the learned dictionary is used to encode the test samples under local constraint to obtain a coefficient matrix with discriminative power. Experimental results show that the proposed MSDL-LCC algorithm framework outperforms some previous state-of-the-art dictionary learning algorithms on the LFW and AR face databases.
KEYWORDS
Face recognition, dictionary learning, virtual sample, locality constrain