Subspace learning with frequency regularizer: Its application to face recognition

Abstract

Subspace learning is an important technique to enhance the discriminative ability of feature representation and reduce the dimension to improve its efficiency. Due to limited training samples and the usual high-dimensional feature, subspace learning always suffers from overfitting problem, which affects its generalization performance. One possible method is to introduce prior information as a regularizer to constrain its solution space. Traditional regularizers are usually designed in spatial domain, which usually make the projection smooth. In this work, we propose a frequency regularizer (FR), which suppresses the high frequency energy so that the smooth priori is incorporated. Two representative supervised subspace methods with frequency regularizer, FR-LDA and FR-SR are introduced and further applied to face recognition problem. Extensive experiments on popular face databases validate the effectiveness and superiority of FR based subspace learning compared to traditional subspace learning methods.