A cognitively-motivated framework for partial face recognition in unconstrained scenarios

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Human face recognition is a complex cognitive process.
• Current automatic face recognition systems struggle with non-ideal image conditions such as occlusion and varying illumination.
• Robust face recognition is crucial for various applications.

Purpose of the Study:

• To develop an automatic face recognition algorithm that addresses performance degradation in non-ideal image acquisition scenarios.
• To propose a novel approach inspired by the global precedent hypothesis of human perception.
• To improve the robustness of face recognition systems.

Main Methods:

• Utilized Scale-Invariant Feature Transform (SIFT) keypoint descriptors.
• Employed a Gaussian Mixture Model (GMM)-based universal background model for descriptor modeling.
• Implemented a hierarchical decision-making process prioritizing holistic information over local analysis.

Main Results:

• The proposed algorithm demonstrated state-of-the-art performance across multiple benchmark face databases (ORL, AR, Extended Yale B).
• Achieved robust face recognition capabilities even under challenging conditions like occlusion and heterogeneous illumination.
• The hierarchical approach proved effective in enhancing recognition accuracy.

Conclusions:

• The novel face recognition approach, inspired by human cognitive mechanisms, offers a significant advancement.
• The algorithm effectively handles non-ideal image acquisition scenarios, outperforming existing methods.
• This work contributes to the development of more reliable and accurate automatic face recognition systems.