Weighted Feature Histogram of Multi-Scale Local Patch Using Multi-Bit Binary Descriptor for Face Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Traditional face recognition relies on single-bit binary descriptors, leading to significant information loss during quantization.
• This loss limits the robustness and accuracy of existing face recognition systems.

Purpose of the Study:

• To propose a novel weighted feature histogram (WFH) method using multi-bit binary descriptors for enhanced face recognition.
• To address the information loss inherent in single-bit descriptors and improve recognition robustness.

Main Methods:

• Multi-scale local patch generation (MSLPG) for comprehensive facial information.
• Multi-bit local binary descriptor learning (MBLBDL) to minimize quantization loss.
• Robust weight learning (RWL) to integrate multi-bit descriptors into a final face representation.

Main Results:

• The proposed WFH method demonstrates superior performance compared to state-of-the-art methods.
• The coupled WFH (C-WFH) variant effectively reduces the modality gap in heterogeneous face recognition.
• Extensive experiments on benchmark datasets validate the efficacy of both WFH and C-WFH.

Conclusions:

• The novel multi-bit descriptor approach significantly enhances face recognition accuracy and robustness.
• WFH and C-WFH offer promising solutions for both standard and heterogeneous face recognition challenges.
• The methods effectively overcome limitations of traditional single-bit descriptor techniques.