Age from Faces in the Deep Learning Revolution.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Facial age estimation is a key area in face analysis, with significant research over two decades.
• Recent advancements in deep learning have dramatically improved performance in various face analysis tasks.
• Existing surveys on face analysis lack recent deep learning approaches for age estimation, with the last survey dating back to 2010.

Purpose of the Study:

• To provide a comprehensive analysis of deep learning methods for facial age estimation proposed in the last six years.
• To review and categorize these methods based on network architecture, learning procedures, datasets, and data augmentation.
• To examine the impact of utilizing auxiliary data like gender, race, and facial expressions on age estimation performance.

Main Methods:

• Systematic review of deep learning-based age estimation approaches from the past six years.
• Analysis of network architectures, training methodologies, and data handling techniques (preprocessing, augmentation).
• Evaluation of studies incorporating additional facial attributes (gender, race, expression) to enhance age estimation.

Main Results:

• Deep learning methods have shown significant performance improvements in facial age estimation.
• Network architecture, learning procedures, and data augmentation strategies critically influence system performance.
• The use of auxiliary data (gender, race, expression) can further boost age estimation accuracy.

Conclusions:

• Deep learning has revolutionized facial age estimation, offering state-of-the-art performance.
• Further research is needed to address open issues and refine existing methods for robust age estimation.
• This review serves as a crucial update, covering the latest deep learning advancements in the field.