Optimized Coefficient Vector and Sparse Representation-Based Classification Method for Face Recognition

Shigang Liu,Ming Jin,Lingjun Li,Yali Peng,Sujuan Hou

doi:10.1109/access.2019.2960928

Abstract

Sparse representation based classification method has led to a wide variety of extensions of representation based methods for face recognition. All of these methods partially reveal that collaborative representation is a crucial factor to make sparse representation based classification powerful for face recognition. The collaborative representation based classification (CRC) methods and corresponding variations have achieved effective results in face recognition. For these methods, we found that the test sample has some relevance with the coefficient vector. For example, nonzero elements in the coefficient vector are associated with the classes which the test sample potentially belong to. Exploiting the relevance may obtain sparser coefficient vector in comparison with the traditional methods. Hence, we propose a novel method in which the test sample is closely involved in the solution procedure of optimal coefficient vector. The classification of the proposed method is performed by checking the minimal residual between the test sample and the collaborative representation with respect the test sample of the selected class, which is similar to that of CRC. The proposed method can intensify the corresponding coefficients in the coefficient vector by exploiting the test sample. Experimental results show that the proposed method does achieve more accurate recognition rate.

Highlights

Sparse representation [1], [2] is based on the theory of compressed sensing, originated in an idea that signals may be expressed at the lowest possible sampling rate
Wright et al [14] proposed the sparse representation based classification (SRC) method, which regards that a test sample can be sparsely coded over a number of training samples and uses the sparse coefficients obtained from the optimization problem to conduct classification of the test sample
After carefully studying extended collaborative representation based classification (CRC) methods, we found that the test sample has some relevance with the coefficient vector

Summary

INTRODUCTION

Sparse representation [1], [2] is based on the theory of compressed sensing, originated in an idea that signals may be expressed at the lowest possible sampling rate. S. Liu et al.: Optimized Coefficient Vector and Sparse Representation-Based Classification Method for Face Recognition. Wright et al [14] proposed the sparse representation based classification (SRC) method, which regards that a test sample can be sparsely coded over a number of training samples and uses the sparse coefficients obtained from the optimization problem to conduct classification of the test sample. Yang et al [21] adopted Gabor local features based SRC to reduce the computational cost These methods all emphasize a point, that is, the sparser the coefficient vector is, the easier it will be to accurately determine the class of the test sample. The proposed method tries to exploit the maximal capability to represent test sample y of all training samples, and enables the optimal coefficient vector to be more discriminative.

DESCRIPTION OF THE PROPOSED METHOD

EXPERIMENT RESULTS

EXPERIMENT ON THE ORL DATABASE

CONCLUSION