A Hybrid Approach for Color Face Recognition Based on Image Quality Using Multiple Color Spaces

Abstract

In this paper, the color face recognition problem is investigated using image quality assessment techniques and multiple color spaces. Image quality is measured using No-Reference Image Quality Assessment (NRIQA) techniques. Color face images are categorized into low, medium, and high-quality face images through the High Low Frequency Index (HLFI) measure. Based on the categorized face images, three feature extraction and classification methods as Scale Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF), and Convolutional Neural Networks (CNN) are applied to face images using RGB, YCbCr, and HSV color spaces to extract the features and then classify the images for face recognition. To enhance color face recognition systems' robustness, a hybrid approach that integrates the aforementioned methods is proposed. Additionally, the proposed system is designed to serve as a secure anti-spoofing mechanism, tested against different attack scenarios, including print attacks, mobile attacks, and high-definition attacks. A comparative analysis that assesses the proposed approach with the state-of-the-art systems using Faces94, ColorFERET, and Replay Attack datasets is presented. The proposed method achieves 96.26%, 100%, and 100% accuracies on ColorFERET, Replay Attack, and Faces94 datasets, respectively. The results of this analysis show that the proposed method outperforms existing methods. The proposed method showcases the potential for more reliable and secure recognition systems.

Keywords

• face recognition
• image quality assessment measures
• color spaces
• feature extraction
• deep learning