Document Type : Research Paper

Authors

• Nabaa Alaa Abdulrazzaq
• Abdulkareem Merhej Radhi

Computer Science Department, College of Science, Al-Nahrain University, Baghdad, Iraq

Abstract

This paper presents a Deep Convolutional Neural Network (DCNN) based facial recognition model that handles illumination, expression, and position variations, among other typical challenges in the area. The model's flexibility and generalizability are enhanced using data augmentation methods for the features extracted from preprocessed face images using CNN. The model was evaluated for performance using five well-recognized datasets: ORL, Yale, Extended Yale B, JAFFE, and LFW. The proposed model attained 97% accuracy on ORL, 93% on Yale, 98% on Extended Yale B, 100% on JAFFE, and 98% on LFW, surpassing current state-of-the-art techniques. To make the model more resilient on smaller datasets such as ORL and JAFFE, data augmentation was performed. On the other hand, Extended Yale B and other more diverse datasets performed well even without augmentation. Also, preprocessing techniques, such as data balance and augmentation, have improved identification abilities, especially in real-world situations like LFW. Overall, this study underscores the power of DCNNs for face recognition and highlights how tailored data augmentation can boost performance across different datasets.

Keywords

• Face recognition
• Deep Learning
• Deep Convolutional Neural Networks
• Data Augmentation
• CNN