Kinship verification via correlation calculation-based multi-task learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Metric learning shows promise in kinship verification.
• Existing methods suffer from information isolation due to limited data types.
• Generative methods are computationally expensive.

Purpose of the Study:

• To propose a novel correlation calculation-based multi-task learning (CCMTL) method for kinship verification.
• To address information isolation and high computational costs in current approaches.
• To leverage correlations between different kinship types for improved metric learning.

Main Methods:

• Developed a multi-task learning framework integrating correlation exploitation with metric learning.
• Investigated spatial distribution relationships to determine correlations among kinship types.
• Designed an efficient algorithm to minimize computational overhead.

Main Results:

• The proposed CCMTL method effectively resolves information isolation.
• CCMTL minimizes computational overhead compared to generative methods.
• Experimental validation on the KinFaceW dataset shows superior or comparable results to existing methods.

Conclusions:

• The CCMTL method offers an efficient and effective solution for kinship verification.
• Leveraging correlations between kinship types enhances metric learning performance.
• This approach provides a promising direction for future research in facial relationship recognition.