Intraspectrum Discrimination and Interspectrum Correlation Analysis Deep Network for Multispectral Face Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Multispectral images offer rich information beyond human vision, making them valuable for face recognition.
• Existing multispectral face recognition methods have not fully exploited intraspectrum and interspectrum information.

Purpose of the Study:

• To propose a novel deep learning approach, the intraspectrum discrimination and interspectrum correlation analysis deep network (IDICN), for enhanced multispectral face recognition.
• To effectively leverage both intraspectrum discriminant information and interspectrum correlation in multispectral face images.

Main Methods:

• The IDICN network divides multiple spectra into spectrum-sets and uses spectrum-set-specific deep convolutional neural networks to extract features.
• A spectrum pooling layer adaptively selects discriminative spectra.
• The network jointly learns nonlinear representations, minimizing intraspectrum Fisher loss and interspectrum discriminant correlation.

Main Results:

• The proposed IDICN approach demonstrates superior performance compared to state-of-the-art methods.
• Experiments were conducted on established multispectral face datasets, including those from Hong Kong Polytechnic University, Carnegie Mellon University, and the University of Western Australia.

Conclusions:

• The IDICN approach effectively enhances multispectral face recognition by analyzing intraspectrum and interspectrum information.
• This method offers a significant advancement in leveraging the full potential of multispectral imaging for biometric identification.