Joint multi-subspace feature learning with singular value decomposition for robust single-sample face recognition
Rong Fei,Jian Zhang,Liling Bo,Heng Zhang,Hongran Li,Ming Li
DOI: https://doi.org/10.1016/j.compeleceng.2024.109085
IF: 4.152
2024-01-20
Computers & Electrical Engineering
Abstract: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 (k-NN) 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.
engineering, electrical & electronic,computer science, interdisciplinary applications, hardware & architecture