Face Feature Weighted Fusion Based on Fuzzy Membership Degree for Video Face Recognition

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• Video face recognition (FR) presents challenges due to variations in pose, illumination, and expression.
• Existing FR methods often struggle with the dynamic and noisy nature of video data.
• Effective fusion of multi-modal information from video frames is crucial for robust performance.

Purpose of the Study:

• To propose a novel video face recognition method that improves accuracy through adaptive fusion of face features from video sequences.
• To theoretically justify the effectiveness of the proposed weighted feature fusion technique by deriving an upper bound for recognition error.
• To develop an optimal weight determination solution for feature fusion using fuzzy logic and image quality assessment.

Main Methods:

• Adaptive fusion of multiple face features extracted from video frames of the same subject.
• Theoretical derivation of an upper bound for recognition error in weighted feature fusion.
• Development of a novel weight determination solution based on fuzzy membership functions and face image quality metrics.
• Evaluation using four public video databases under realistic conditions.

Main Results:

• The proposed method demonstrates significant improvements in video face recognition accuracy.
• Theoretical analysis provides a strong foundation for the effectiveness of the weighted fusion approach.
• The fuzzy-based weight determination effectively optimizes feature fusion for enhanced performance.
• Experimental results validate the method's efficacy in real-world video FR scenarios.

Conclusions:

• The proposed adaptive fusion method offers a simple yet effective solution for video face recognition.
• The theoretical framework and practical implementation highlight the robustness and accuracy of the approach.
• This method has the potential to advance the field of biometric security and surveillance systems.