Distribution Cognisant Loss for Cross-Database Facial Age Estimation With Sensitivity Analysis

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Facial age estimation often relies on intra-database protocols, limiting real-world applicability.
• Cross-database age estimation faces challenges due to differing image characteristics between training and testing sets.

Purpose of the Study:

• To address subjective-exclusive cross-database age estimation challenges.
• To propose a robust distribution learning framework for age estimation.
• To enhance the generalization capability of facial age estimation models.

Main Methods:

• Formulating age estimation as a distribution learning problem.
• Developing a novel loss function for robust distribution comparison.
• Introducing a subject-exclusive cross-database evaluation protocol.
• Compiling a new balanced large-scale age estimation database.

Main Results:

• The proposed loss function demonstrates superior performance in cross-database age estimation.
• The new evaluation protocol effectively assesses generalization capabilities.
• Experimental results show the approach outperforms state-of-the-art methods in both intra-database and cross-database settings.

Conclusions:

• The proposed distribution learning framework and loss function significantly improve facial age estimation across different databases.
• The study highlights the importance of subject-exclusive evaluation for assessing real-world performance.
• Identified open problems offer future research directions for robust age estimation systems.