Relative Attribute SVM+ Learning for Age Estimation.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Human experts utilize privileged information (facial attributes) for accurate age estimation.
• Automatic age estimation models lack this privileged information in test images, limiting generalizability.
• The learning using privileged information (LUPI) framework offers a potential solution.

Purpose of the Study:

• To investigate the exploitation of asymmetric information for enhancing automatic age estimation.
• To develop a novel method, relative attribute SVM+ (raSVM+), within the LUPI framework.
• To improve the generalizability and performance of age estimation models.

Main Methods:

• Defined relative attributes for support vector machine (SVM+) to leverage privileged information.
• Implemented the raSVM+ approach, integrating privileged information for outlier separation and error manipulation during training.
• Utilized the learning using privileged information (LUPI) framework.

Main Results:

• raSVM+ demonstrated superiority over state-of-the-art algorithms on benchmark datasets (FG-NET, craniofacial longitudinal morphological face aging).
• Achieved a mean absolute error of 4.07 on the FG-NET dataset.
• The method effectively guides the trained predictor toward a more generalizable solution.

Conclusions:

• raSVM+ is a promising development for improving automatic age estimation.
• Exploiting privileged information via LUPI significantly enhances model performance and generalizability.
• The proposed method offers a robust approach to address the challenge of missing privileged information in test data.