Cross-Generation Kinship Verification with Sparse Discriminative Metric

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Kinship verification is crucial for applications like personal album organization, missing person investigations, and forensic analysis.
• Verifying family pairs with generational gaps (e.g., father-son) is challenging due to age differences and identity variations.

Purpose of the Study:

• To develop an effective kinship verification model that mitigates age and identity divergences.
• To enhance the accuracy of identifying familial relationships across generations.

Main Methods:

• Proposed a 'towards-young' cross-generation model utilizing a conditional generative model to create an intermediate domain.
• Employed deep architectures with a novel Sparse Discriminative Metric Loss (SDM-Loss) to extract discriminative features.
• The SDM-Loss incorporates both positive and negative kinship information for improved feature learning.

Main Results:

• The proposed model demonstrated superior performance on kinship verification benchmarks.
• Achieved state-of-the-art results compared to existing kinship verification methods.
• Effectively addressed challenges posed by age gaps and identity variations in kinship verification.

Conclusions:

• The 'towards-young' cross-generation model offers a promising solution for accurate kinship verification.
• The integration of conditional generative models and SDM-Loss significantly improves performance in cross-generational kinship identification.
• This method provides a robust approach for real-world applications requiring reliable familial relationship verification.