SELM: Siamese extreme learning machine with application to face biometrics

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Extreme Learning Machine (ELM) is a fast classification method but struggles with face verification due to its single-input structure.
• Existing ELM adaptations for face verification use concatenated inputs, doubling computational cost and hindering metric learning.

Purpose of the Study:

• To introduce a Siamese Extreme Learning Machine (SELM) designed for parallel processing of dual-input data streams.
• To develop a Gender-Ethnicity-dependent triplet feature for improved facial feature extraction across demographic groups.

Main Methods:

• Proposed SELM architecture with a dual-stream Siamese network for parallel data processing.
• Introduced a novel Gender-Ethnicity-dependent triplet feature for specialized demographic group analysis.
• Conducted comparative experiments against ELM and Deep Convolutional Neural Network (DCNN).

Main Results:

• The proposed feature achieved $$ accuracy and $$ AUC.
• SELM combined with the proposed feature yielded $$ accuracy and $$ AUC.
• SELM demonstrated superior performance over DCNN and standard ELM methods.

Conclusions:

• SELM effectively addresses the limitations of ELM in face verification tasks.
• The Gender-Ethnicity-dependent triplet feature significantly enhances facial recognition accuracy.
• SELM offers a more efficient and accurate solution for face verification compared to existing methods.