Multi-Block Color-Binarized Statistical Images for Single-Sample Face Recognition.

Area of Science:

• Computer Vision
• Biometrics
• Image Processing

Background:

• Single-Sample Face Recognition (SSFR) presents significant challenges due to limited training data per individual.
• Unconstrained environments introduce variability in facial expression, pose, lighting, and occlusion, complicating accurate identification.
• Existing SSFR methods often struggle with these real-world variations.

Purpose of the Study:

• To introduce and evaluate an original method, Multi-Block Color-Binarized Statistical Image Features (MB-C-BSIF), for robust SSFR.
• To demonstrate the efficacy of MB-C-BSIF in handling variations common in unconstrained face recognition scenarios.
• To provide an efficient alternative to computationally intensive methods like deep learning for real-time applications.

Main Methods:

• The MB-C-BSIF method decomposes facial images into RGB channels.
• Each channel is divided into non-overlapping blocks to extract local, regional, global, and textured-color features.
• Classification is performed using a K-nearest neighbors (K-NN) classifier with distance measurements.

Main Results:

• MB-C-BSIF achieved high classification accuracies of 96.17% and 99% on the AR database (Protocols I and II).
• The method attained 38.01% accuracy on the challenging Labeled Faces in the Wild (LFW) database.
• Experimental results on AR and LFW databases show superior performance compared to state-of-the-art methods, especially under varying conditions.

Conclusions:

• The MB-C-BSIF method offers a highly effective solution for Single-Sample Face Recognition in unconstrained environments.
• Its reliance on simple image processing operations ensures low computational cost, making it suitable for real-time identification.
• MB-C-BSIF demonstrates significant improvements over existing methods in handling facial variations and occlusions.