Illumination invariant face recognition using near-infrared images

Area of Science:

• Computer Science
• Biometrics
• Image Processing

Background:

• Current face recognition systems struggle with accuracy due to varying illumination conditions.
• Existing academic and commercial systems are compromised in indoor, cooperative-user applications by environmental lighting changes.

Purpose of the Study:

• To develop a novel solution for illumination-invariant face recognition in indoor, cooperative-user settings.
• To present an active near-infrared (NIR) imaging system for consistent face image acquisition.
• To create an accurate and fast face recognition system that addresses challenges like eyeglasses.

Main Methods:

• Utilized an active near-infrared (NIR) imaging system to capture face images independent of visible light.
• Employed local binary pattern (LBP) features to compensate for monotonic gray tone transforms, achieving an illumination-invariant face representation.
• Applied statistical learning algorithms to extract discriminative features from LBP features for a high-accuracy face matching engine.
• Developed a method to mitigate specular reflections from NIR lights on eyeglasses.

Main Results:

• The NIR imaging system produces good quality face images across different visible light conditions.
• The LBP-based representation effectively compensates for illumination variations.
• The developed face recognition system demonstrates accurate and fast performance.
• The system successfully handles challenges posed by eyeglasses.

Conclusions:

• The proposed active NIR imaging system and LBP-based feature representation offer a robust solution for illumination-invariant face recognition.
• The developed system achieves high accuracy and speed, even in the presence of challenging factors like eyeglasses.
• This approach significantly improves face recognition reliability in indoor environments with variable lighting.