Single Sample Face Recognition under Varying Illumination via QRCP Decomposition.

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Single Sample Face Recognition (SSFR) is challenging due to variations in illumination.
• Existing methods struggle to model intraclass variations caused by illumination changes.

Purpose of the Study:

• To develop a robust high-frequency facial feature for SSFR under varying illumination.
• To propose a high-frequency based sparse representation classification model to address illumination variations.

Main Methods:

• Proposed QRCP-face (and its normalized and adaptive variants) to capture intrinsic face surface features and illumination intensities.
• Developed a General High-Frequency based Sparse Representation (GHSP) model, including SHSP and UHSP approaches.
• Validated methods on multiple benchmark face databases (Extended Yale B, CMU PIE, AR, LFW) and a custom dataset.

Main Results:

• The proposed QRCP-face effectively represents facial features and handles illumination variations.
• The GHSP model demonstrates superior performance in modeling intraclass variations.
• Experimental results show the proposed methods outperform previous approaches in SSFR under varying illumination.

Conclusions:

• The novel high-frequency facial feature and sparse representation effectively tackle SSFR under varying illumination.
• The proposed methods offer a significant advancement for face recognition systems operating in uncontrolled lighting conditions.