Title: Cross-modal correlation feature extraction with orthogonality redundancy reduce and discriminant structure constraint

Authors: Qianjin Zhao; Xinrui Ping; Shuzhi Su

Addresses: School of Mathematics and Big Data, Anhui University of Science and Technology, Huainan 232001, China ' School of Mathematics and Big Data, Anhui University of Science and Technology, Huainan 232001, China ' School of Computer Science and Engineering, Anhui University of Science and Technology, Huainan 232001, China

Abstract: Canonical correlation analysis (CCA) is a classic feature extraction method that is widely used in the field of pattern recognition. Its goal is to learn correlation projection directions to maximise the correlation between the two sets of variables, but it does not take into consideration the class label information among samples and the within-modal redundancy information from the correlation projection directions. To this end, a novel method of class label embedding orthogonal correlation feature extraction method is proposed in this paper. The label-guide discriminant structure information is deeply embedded in correlative analytical theories for improving discrimination of correlation features, and within-modal orthogonality constraints are added to further reduce the projection redundancy of correlation features. Several validation experiments on the GT, Umist and YALE datasets are designed, which demonstrates that DOCCA method is superior to other feature extraction methods and achieves state-of-the-art performance. This method provides a new solution to pattern recognition.

Keywords: feature extraction; correlation analysis theory; discriminative subspace learning; orthogonality redundancy reduce.

DOI: 10.1504/IJCSE.2023.129151

International Journal of Computational Science and Engineering, 2023 Vol.26 No.1, pp.101 - 109

Received: 02 Jul 2021
Accepted: 03 Nov 2021
Published online: 23 Feb 2023 *