Knowledge-based tensor subspace analysis system for kinship verification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biometrics

Background:

• Existing automatic kinship verification methods often struggle with simultaneously learning facial and kinship features, leading to weaker models.
• There is a need for robust methods that can effectively fuse diverse features for accurate kinship verification.

Purpose of the Study:

• To develop a knowledge-based tensor similarity extraction framework for automatic facial kinship verification.
• To bridge the gap in current methods by integrating pre-trained multi-view models and diverse feature sets.

Main Methods:

• Utilized four pre-trained networks (VGG-Face, VGG-F, VGG-M, VGG-S) within a knowledge-based tensor framework.
• Fused deep face and general features using a tensor design for comprehensive kinship cue understanding.
• Employed Tensor Cross-view Quadratic Exponential Discriminant Analysis with margin maximization for learning representations.

Main Results:

• Successfully reduced the intra-class variability using the Whitened Common Covariance (WCCN) metric.
• Demonstrated significant improvements in distinguishing between same-family and different-family distributions.
• Achieved state-of-the-art performance across four challenging kinship verification datasets.

Conclusions:

• The proposed knowledge-based tensor framework offers an effective approach to automatic facial kinship verification.
• The method successfully integrates diverse features, enhancing the understanding of kinship cues.
• The system addresses explainability of black-box models and challenges in ubiquitous face recognition.