Abstract

Linear regression has shown an effective tool for face recognition in recent years. Most existing linear regression based methods are devised for grayscale image based face recognition and fail to exploit the color information of color face images. To extend linear regression for color images, we propose a novel color face recognition method by formulating the color face recognition problem as a quaternion linear regression model. The proposed quaternion linear regression classification (QLRC) algorithm models each color facial image as a quaternion signal and codes multiple channels of each query color image in a holistic manner. Thus, the correlation among distinct channels of each color image is well preserved and leveraged by QLRC to further improve the recognition performance. To further improve QLRC, we propose a quaternion collaborative representation optimized classifier (QCROC) which integrates QLRC and quaternion collaborative representation based classifier into a unified framework. The experiments on benchmark datasets demonstrate the efficacy of the proposed approaches for color face recognition.

Highlights

  • Face recognition (FR) has been a hot research topic in pattern recognition for decades due to its wide applications in reality [1]–[5]

  • To further improve quaternion linear regression classification (QLRC), we develop a quaternion collaborative representation optimized classifier (QCROC) by integrating QLRC and quaternion collaborative representation (QCR) into a unified framework

  • We show that QCROC can take advantage of both merits of QLRC and QCR and outperforms them for face recognition

Read more

Summary

Introduction

Face recognition (FR) has been a hot research topic in pattern recognition for decades due to its wide applications in reality [1]–[5]. In the past two decades, a variety of FR methods have been proposed in the literatures [3], [4], [6], [7]. Deep learning based classifiers have shown impressive performance in many visual tasks including face recognition [8]–[10]. They often rely heavily on a large number of training samples, which are often limited and the acquisition of labeled data is generally very laborious in reality and difficult in some cases [11].

Objectives
Methods
Findings
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call