Efficient Group-n Encoding and Decoding for Facial Age Estimation.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Aging is a complex, non-stationary process with inherent randomness.
• Accurate age estimation is crucial for various applications.

Purpose of the Study:

• To propose a novel method for improving age estimation accuracy.
• To explore the relationship between an individual's real age and adjacent age groups.

Main Methods:

• Developed the age group-n encoding (AGEn) method, grouping adjacent ages into classes.
• Transformed age estimation into multiple binary classification sub-problems.
• Employed deep Convolutional Neural Networks (CNNs) with multiple classifiers and a Local Age Decoding (LAD) strategy.
• Incorporated a penalty factor into the objective function to address data imbalance.

Main Results:

• The proposed AGEn method achieved state-of-the-art performance on FG-NET, MORPH II, CACD, and Chalearn LAP 2015 datasets.
• The Local Age Decoding strategy accelerated the age prediction process.
• The penalty factor effectively alleviated data imbalance issues for classifiers.

Conclusions:

• The AGEn method offers a robust and accurate approach to age estimation.
• The study demonstrates the effectiveness of transforming age estimation into binary classification sub-problems.
• The proposed techniques significantly advance the field of facial age estimation.