Effective face recognition using dual linear collaborative discriminant regression classification algorithm

In recent decades, face recognition is an attractive and emerging research area in computer vision and pattern recognition applications. Still, facial recognition is a challenging task due to the following factors; different face expressions, posture, wearing of scarves/glasses, and illumination conditions. To overcome the aforementioned factors, a new Dual Linear Collaborative Discriminant Regression Classification (DLCDRC) algorithm is introduced in this research paper. The proposed DLCDRC algorithm contains dual steps such as deep loss function, and combined distance metric. The deep loss function includes inter-class and intraclass loss functions, which are used to decrease inter and intraclass variations in Within Class Reconstruction Error (WCRE), and Collaborative between Class Reconstruction Error (CBCRE). Additionally, the combined distance metric is used to construct CBCRE and WCRE with minimum reconstruction error values that efficiently improve the face recognition performance. By inspecting the resulting phase, the DLCDRC algorithm achieved 92.47%, 96.39%, and 98.86% of classification accuracy on YALE B, ORL, and extended YALE B face datasets. The proposed DLCDRC method has the improvement of 1.85% in Extended YALE B dataset, 3.05% improvement in ORL dataset, and improvement of 2.4% in YALE B dataset. The obtained simulation results are better compared to the existing linear regression algorithms on three benchmark datasets.

Automated summary

The research introduces a new Dual Linear Collaborative Discriminant Regression Classification (DLCDRC) algorithm, which enhances face recognition performance through deep loss function and combined distance metric steps, achieving high classification accuracy on YALE B, ORL, and extended YALE B face datasets.