A Benchmark and Comparative Study of Video-Based Face Recognition on COX Face Database.

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Face recognition research is dominated by still image analysis, with limited focus on video-based methods.
• Real-world applications necessitate distinct approaches for Video-to-Still (V2S), Still-to-Video (S2V), and Video-to-Video (V2V) recognition.
• Existing datasets and evaluation protocols inadequately address all three video-based face recognition scenarios.

Purpose of the Study:

• To address the lack of comprehensive benchmarks for video-based face recognition.
• To facilitate research across V2S, S2V, and V2V scenarios.
• To provide a new large-scale dataset and evaluation framework.

Main Methods:

• Collection and release of the COX Face Database, a large-scale still/video face dataset.
• Review and comparative benchmarking of existing set-based methods on the new database.
• Proposal and experimental validation of a novel Point-to-Set Correlation Learning (PSCL) method.

Main Results:

• The COX Face DB supports benchmarking for V2S, S2V, and V2V face recognition scenarios.
• Comparative analysis highlights the performance variations of existing set-based methods.
• The proposed PSCL method demonstrates promise as a baseline for V2S and S2V recognition.

Conclusions:

• Video-based face recognition requires further research and development.
• The COX Face DB serves as a valuable benchmark for evaluating video-based face recognition techniques.
• The PSCL method offers a potential advancement for specific video-based face recognition tasks.