Multilayer surface albedo for face recognition with reference images in bad lighting conditions

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• Illumination variations pose significant challenges in face recognition.
• Existing methods often struggle with robustness due to detrimental sharp features in surface albedo.
• Need for improved face recognition models under adverse lighting conditions.

Purpose of the Study:

• To propose a novel multilayer surface albedo (MLSA) model for robust face recognition.
• To enhance the representation of surface albedo by decomposing it into detailed layers.
• To improve recognition performance, particularly when reference images have poor lighting.

Main Methods:

• Developed a multilayer surface albedo (MLSA) model.
• Decomposed surface albedo into a linear sum of detailed layers to separate features by scale.
• Introduced a criterion function to select layer weights using an independent training set.

Main Results:

• The MLSA model demonstrated effectiveness against controlled and uncontrolled illumination variations.
• The model showed robustness against other complex variations like expression, pose, and occlusion.
• Experiments on benchmark datasets confirmed good receiver operating characteristic curves and statistical discriminating capability.

Conclusions:

• The refined surface albedo significantly improves face recognition performance.
• MLSA is particularly effective for face recognition with reference images under poor lighting conditions.
• The proposed model offers a robust solution for face recognition in challenging environments.