Joint multi-subspace feature learning with singular value decomposition for robust single-sample face recognition

doi:10.1016/j.compeleceng.2024.109085

Joint multi-subspace feature learning with singular value decomposition for robust single-sample face recognition

2024-03-01

1

PRE

AI

DOI：10.1016/j.compeleceng.2024.109085

Original

Original request for help

Save

Abstract

En 中文

Single -sample face recognition remains a significant challenge due to the difficulty in extracting discriminative features using only one facial image per individual in practical applications. In light of this, the paper introduces a new method for learning multiple subspace features using singular value decomposition (SVD). Specifically, we divide each facial image into two symmetrical halves to increase intra-class diversity. The SVD is subsequently applied to each half to create distinct geometric view subspaces. Then, discriminative features are learnt through performing 2 -dimension linear discriminant analysis (2DLDA) in each subspace. Lastly, the identification of faces is accomplished by using a k -nearest neighbour classifier (kNN) within each subspace followed by a majority voting strategy. The method proposed is extensively verified through rigorous experiments carried out on different databases such as CUHK, Extended Yale B, FRGCv2 and AR. The experimental outcomes unambiguously show that our approach consistently achieves competitive performance when compared to other methods.

Keywords:

Single-sample face recognition

Singular value decomposition

Multiple subspace features

Facial base image

Joint multi-subspace feature learning with singular value decomposition for robust single-sample face recognition

Abstract

Journal

Organization