High-resolution face verification using pore-scale facial features.

Area of Science:

• Computer Vision
• Biometrics
• Pattern Recognition

Background:

• Traditional face recognition methods struggle with pose variations and alignment errors.
• High-resolution (HR) face data is increasingly available but underutilized in face recognition.
• Existing methods often require multiple images per subject or struggle with single-gallery scenarios.

Purpose of the Study:

• To develop a pose-invariant face verification method robust to alignment errors.
• To leverage HR face information for improved face recognition accuracy.
• To address limitations of current methods in handling expression and pose variations with limited gallery data.

Main Methods:

• Proposed a novel pose-invariant face-verification method utilizing HR information and pore-scale facial features.
• Devised a new keypoint descriptor, pore-Principal Component Analysis (PCA)-Scale Invariant Feature Transform (PPCASIFT), adapted from PCA-SIFT.
• Implemented a robust-fitting scheme for matching pore-scale features for verification.

Main Results:

• The proposed PPCASIFT method extracts distinctive pore-scale facial features.
• The robust-fitting scheme effectively verifies faces based on matched pore-scale features.
• Achieved superior performance compared to standard verification methods with only one frontal-view gallery image per subject.

Conclusions:

• The HR-based pore-scale feature method offers robust face verification against pose and expression variations.
• The approach overcomes limitations of alignment-dependent and feature-based methods.
• Demonstrated excellent accuracy in challenging face verification scenarios.