Hyperspectral Face Recognition with Adaptive and Parallel SVMs in Partially Hidden Face Scenarios

Area of Science:

• Computer Vision
• Biometrics
• Spectroscopy

Background:

• Masked face recognition is a growing challenge.
• Hyperspectral imaging offers unique spectral information for discrimination.
• Traditional methods struggle with occluded faces.

Purpose of the Study:

• To develop a novel algorithm for masked face recognition using hyperspectral imaging.
• To extract and classify facial spectral-features from unoccluded regions.
• To optimize the trade-off between recognition accuracy and compression ratio.

Main Methods:

• Applied computer vision techniques, specifically Histogram of Oriented Gradients, to hyperspectral images.
• Extracted facial spectral-features from regions of interest.
• Trained parallel Support Vector Machines with custom kernels (cosine similarity, Euclidean distance) on the UWA-HSFD dataset.

Main Results:

• Successfully extracted discriminative facial spectral-features from visible facial areas.
• Achieved an optimal balance between recognition accuracy and compression ratio.
• Demonstrated the effectiveness of the algorithm for masked face recognition.

Conclusions:

• Hyperspectral imaging combined with advanced algorithms can overcome face mask occlusions.
• The proposed method provides a robust approach for masked face recognition.
• The algorithm effectively utilizes spectral information from unoccluded facial regions.