Robust Single-Sample Face Recognition by Sparsity-Driven Sub-Dictionary Learning Using Deep Features.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Face recognition with a single reference image per subject (SSPP) is difficult, especially with large subject galleries and unconstrained image acquisition.
• Unconstrained conditions include variations in illumination, pose, facial expression, partial occlusions, and low resolution, significantly increasing problem complexity.

Purpose of the Study:

• To address the challenging Single Sample Per Person (SSPP) face recognition problem in large-scale, unconstrained datasets.
• To develop a robust method capable of overcoming common image variability issues encountered in real-world scenarios.

Main Methods:

• The proposed technique, k-LiMapS, utilizes sparse dictionary learning (method of optimal direction) and iterative ℓ 0-norm minimization.
• It operates on deep-learned features, augmented to enhance image variability.
• The method is evaluated on large-scale, unconstrained datasets like LFW and AR, including tests on very low-resolution images.

Main Results:

• k-LiMapS demonstrates effectiveness in handling illumination, pose, expression, occlusions, and low-resolution challenges.
• Experiments on the LFW dataset with up to 1680 subjects show superior performance.
• Tests on very low-resolution images (down to 8x8 pixels) and the AR dataset with disguises confirm the method's robustness.

Conclusions:

• The k-LiMapS method significantly outperforms state-of-the-art approaches for SSPP face recognition under challenging, unconstrained conditions.
• The technique offers a robust solution for real-world face recognition applications where image quality and consistency are compromised.