Graphical Representation for Heterogeneous Face Recognition.

Area of Science:

• Computer Science
• Biometrics
• Image Processing

Background:

• Heterogeneous face recognition (HFR) is crucial for biometrics but challenging due to difficulties in representing diverse image sources homogeneously.
• Current HFR methods often neglect spatial information or use complex transformations, hindering performance.

Purpose of the Study:

• To propose a novel graphical representation-based HFR method (G-HFR) that addresses limitations of existing approaches.
• To enhance the accuracy and robustness of face identification across heterogeneous image modalities.

Main Methods:

• Employing Markov networks to represent individual heterogeneous image patches, preserving spatial relationships.
• Introducing a coupled representation similarity metric (CRSM) for comparing graphical representations.
• Evaluating the method across various HFR scenarios including sketches and infrared imagery.

Main Results:

• The proposed G-HFR method demonstrates superior performance compared to state-of-the-art techniques.
• Experiments validate the effectiveness of the graphical representation and similarity metric in HFR tasks.
• The method shows robustness across diverse HFR scenarios.

Conclusions:

• The novel graphical representation approach effectively handles spatial information in HFR.
• G-HFR offers a promising solution for accurate and reliable heterogeneous face identification.
• This work advances the field of biometrics by improving cross-modal face matching capabilities.