Kernel difference maximisation-based sparse representation for more accurate face recognition

Abstract

Most methods for sparse representation are designed to be used in the original space. However, their performance is not always satisfactory especially when training samples are limited. According to the previous studies, more information can be obtained from samples in the feature space than those in the original space. The authors propose a novel kernel difference maximisation-based sparse representation method, and its remarkable performance in face recognition is demonstrated by the experiments. The proposed method converts all the samples into the feature space, and a test sample can be denoted as a representation with all the training samples’ linear combinations. Besides, a novel solution scheme for sparse representation is utilised to obtain the [inline-formula] regularisation-based sparse solution. Finally, the classification of the test sample can be easily judged according to the representation result. The representation results of test samples from different classes obtained by their method are very different, making the classification of test samples easier. Besides, the proposed method is simpler than the related methods and does not require dictionary learning.

Keywords

• feature extraction
• learning (artificial intelligence)
• face recognition
• image representation
• image classification
• optimisation
• l(2) regularisation-based sparse solution
• training sample linear combinations
• test sample classification
• novel kernel difference maximisation-based sparse representation method
• feature space
• face recognition