Auxiliary Demographic Information Assisted Age Estimation With Cascaded Structure

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Accurate age estimation is complex due to intrinsic and extrinsic individual variations.
• Existing age estimation methods face challenges in achieving high precision.

Purpose of the Study:

• To propose novel cascaded structure frameworks for enhanced age estimation using convolutional neural networks (CNNs).
• To investigate the utility of auxiliary demographic information (gender, race) in improving age prediction accuracy.

Main Methods:

• Development of five cascaded CNN frameworks, each comprising a parent network and subnetworks.
• Integration of auxiliary demographic data (gender, race) to guide network learning.
• Application of Gaussian process regression on extracted features to further boost performance.

Main Results:

• Achieved superior performance on MORPH II and CACD datasets compared to state-of-the-art methods.
• Significantly reduced the mean absolute error (MAE) from 3.63 to 2.93 years on the MORPH II dataset.
• Demonstrated the effectiveness of cascaded structures and demographic information in age estimation.

Conclusions:

• The proposed cascaded CNN frameworks offer a robust approach to age estimation.
• Auxiliary demographic information plays a crucial role in enhancing prediction accuracy.
• The method shows significant potential for real-world applications requiring precise age estimation.