Recognizing blurred, nonfrontal, illumination, and expression variant partially occluded faces

Area of Science:

• Computer Vision
• Biometrics
• Image Processing

Background:

• Face recognition systems struggle with variations in motion blur, pose, illumination, expression, and occlusion.
• Existing methods often require multiple images or struggle with single-gallery image scenarios.

Purpose of the Study:

• To develop a robust face recognition method that addresses space-varying motion blur and pose variations.
• To enable accurate face recognition using only a single image per subject in the gallery.

Main Methods:

• Estimating motion blur from alpha matte and automatically generating trimaps.
• Synthesizing frontal gallery images to match probe pose variations.
• Modeling probe images using blurred illumination basis images and sparse occlusion representation.
• Employing a recognition metric that leverages the sparsity of occluded pixels.

Main Results:

• The method effectively estimates motion blur from alpha mattes.
• Pose variations are compensated for by synthesizing images.
• Illumination, expression, and occlusion are handled through a linear combination model.
• Performance validated on both synthetic and real face datasets.

Conclusions:

• The proposed approach significantly enhances face recognition accuracy under challenging conditions.
• It offers a robust solution for single-gallery image face recognition with motion blur and pose variations.