Facial Age Estimation With Age Difference.

Area of Science:

• Computer Vision
• Pattern Recognition
• Machine Learning

Background:

• Accurate age estimation from facial images is a significant challenge.
• Existing methods often require large, labeled datasets, limiting the use of abundant unlabeled social media data.
• Weakly labeled data, such as age differences between image pairs of the same person, offer a valuable alternative.

Purpose of the Study:

• To develop a novel deep learning scheme for improved facial age estimation.
• To leverage weakly labeled data, specifically age differences, for training.
• To overcome the limitations of traditional methods requiring extensive age-labeled datasets.

Main Methods:

• A deep convolutional neural network (CNN) based learning scheme was proposed.
• Kullback-Leibler divergence was used to embed age difference information from image pairs.
• Adaptive entropy and cross-entropy losses were applied to guide the network's learning process.
• A new dataset of over 100,000 face images with timestamps and identities was created.

Main Results:

• The proposed age difference learning system demonstrated significant advantages.
• State-of-the-art performance was achieved on two aging face databases.
• The method effectively learns age progression from age difference information alone.

Conclusions:

• The novel learning scheme successfully utilizes weakly labeled facial data for age estimation.
• Deep convolutional neural networks combined with age difference learning offer a powerful approach.
• This method enhances the utilization of large-scale, unlabeled facial image datasets.