Efficient detection of occlusion prior to robust face recognition

Area of Science:

• Computer Vision
• Biometrics
• Pattern Recognition

Background:

• Face recognition research has focused on pose, illumination, and expression variations.
• Facial occlusions, such as sunglasses, hats, scarves, and beards, significantly degrade recognition system performance in uncontrolled environments like video surveillance.
• Occlusion-related challenges in face recognition have received less research attention compared to other variations.

Purpose of the Study:

• To explore and improve face recognition performance in the presence of partial facial occlusions.
• To develop an efficient method for face recognition in real-world scenarios with occlusions (e.g., sunglasses, scarves).

Main Methods:

• Proposed an efficient approach involving two main steps: occlusion analysis and recognition on non-occluded regions.
• Utilized selective local Gabor binary patterns (LGBPHS) for feature extraction from unoccluded facial areas.
• Developed a system to analyze the presence of potential occlusions on a face.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art methods (KLD-LGBPHS, S-LNMF, OA-LBP, RSC).
• Achieved significant results even when evaluated under varying illumination conditions and extreme facial expressions.
• Effectively recognized faces by focusing on non-occluded facial regions.

Conclusions:

• The developed approach offers an effective solution for face recognition under partial occlusions.
• The method shows robustness and high performance in challenging real-world scenarios.
• Highlights the importance of addressing occlusion issues for practical face recognition systems.