One shot learning approach for cross spectrum periocular verification.

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• The COVID-19 pandemic increased demand for periocular biometrics in surveillance.
• Matching cross-spectrum periocular images is challenging due to illumination variations and limited/imbalanced datasets.

Purpose of the Study:

• To propose a Siamese Convolutional Neural Network (CNN) architecture for effective periocular verification.
• To address challenges in cross-spectrum periocular image matching, including illumination variations and data scarcity.

Main Methods:

• Proposed a Siamese CNN architecture based on one-shot classification principles.
• Employed Contrast Limited Adaptive Histogram Equalization (CLAHE) for image preprocessing to handle illumination variations.
• Evaluated the model on IMP, UTIRIS, and PolyU databases using Binary Cross-Entropy, Hinge, Contrastive, and Triplet loss functions.

Main Results:

• The Siamese CNN with triplet loss demonstrated superior performance in periocular verification.
• The proposed method achieved state-of-the-art results for cross, mono, and multi-spectral periocular image matching.
• One-shot classification effectively mitigated the need for large, balanced datasets.

Conclusions:

• The proposed Siamese CNN model offers a robust solution for periocular verification, particularly in challenging cross-spectrum scenarios.
• The one-shot learning approach effectively handles data limitations and class imbalance issues.
• This research advances the field of periocular biometrics for surveillance applications.