Kinship Verification Based on Cross-Generation Feature Interaction Learning.

Area of Science:

• Computer Vision
• Biometrics

Background:

• Kinship verification from facial images is an emerging and challenging computer vision task.
• Existing methods often rely on traditional similarity measures, which may not fully capture complex familial relationships.

Purpose of the Study:

• To propose a novel Cross-Generation Feature Interaction Learning (CFIL) framework for robust kinship verification.
• To improve the accuracy and comprehensiveness of facial kinship verification by exploring cross-generation relationships.

Main Methods:

• Developed a collaborative weighting strategy to extract features from parent-child image pairs.
• Integrated similarity learning and feature extraction into a unified deep Convolutional Neural Network (CNN) architecture.
• Employed interior auxiliary weights derived from similarity calculations to learn holistic image features.

Main Results:

• The proposed CFIL framework demonstrated superior performance compared to state-of-the-art methods.
• Experiments validated the efficiency and effectiveness of the integrated learning approach.
• The model successfully captured both local and non-local features for comprehensive semantic representation.

Conclusions:

• The CFIL framework offers a robust and effective solution for kinship verification from facial images.
• Integrating similarity learning with feature extraction enhances the model's ability to preserve correlation knowledge.
• This approach advances the field of facial biometrics and computer vision applications.