Facial age recognition based on deep manifold learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Facial age recognition is crucial for real-world applications.
• Current deep learning methods for facial age recognition often extract redundant features, hindering performance.
• High-dimensional facial data presents challenges for accurate age estimation.

Purpose of the Study:

• To propose a novel Deep Manifold Learning (DML) approach for effective facial age recognition.
• To enhance the accuracy of age estimation by selecting relevant age-related features.
• To address the issue of redundant feature extraction in deep learning-based facial age recognition.

Main Methods:

• Deep learning was employed to extract high-dimensional facial features.
• Manifold learning was utilized to select age-related features from the extracted high-dimensional data.
• The proposed DML method was validated on the MORPH and FG-NET datasets.

Main Results:

• The Deep Manifold Learning (DML) method achieved a Mean Absolute Error (MAE) of 1.60 on the MORPH dataset.
• The DML method achieved a Mean Absolute Error (MAE) of 2.48 on the FG-NET dataset.
• DML demonstrated significant improvements in accuracy compared to state-of-the-art facial age recognition methods.

Conclusions:

• Deep Manifold Learning (DML) effectively reduces redundant features for improved facial age recognition.
• The proposed DML method offers superior performance in age estimation tasks.
• DML represents a promising advancement in the field of facial age recognition.