Summary

The aim of this project is to make a research and find a solution for age classification based on face recognition. Based on face images collected from the camera, the ML model should classify faces into one of the specific age ranges.

Sources Review

The following articles were analysed:

1. Gupta, S.K., Nain, N. Review: Single attribute and multi attribute facial gender and age estimation. Multimed Tools Appl (2022)

2. ELKarazle, K.; Raman, V.; Then, P. Facial Age Estimation Using Machine Learning Techniques: An Overview. Big Data Cogn. Comput. 2022, 6, 128.

Research Results

The result of facial age estimation can be either age range (classification problem) or exact age value (regression problem).
Main challenges in building efficient age prediction systems are:

1. Real-life factors of face imaging such as resolution, sharpness, illumination, expression, occlusion, profile, frontal view, constraint environment, unconstraint environment, longitudinal, race, hair, scale, etc.
2. Datasets creation

Existing datasets usually divided into the following types:

• controlled (prepared in controlled environment with limited variability). Examples: FG-NET, MORPH, VADANA

• uncontrolled (prepared in real-life with different variability). Examples: LFW, IMDB-WIKI, ChaLearn Looking at People (LAP), Specs on Face (SoF), MSU LFW+, YGA

Steps for dataset creation:

1. Image of faces collection
2. Pre-processing
   • cropping
   • rotating
   • aligning
   • augmenting

Evaluation metrics used:

• accuracy: \((true positive + true negative)/(total test samples)\), and

• Mean Absolute Error (MAE): mean value of the absolute differences between predicted age and real age (ground truth) of test samples

Steps of model building

1. Feature extraction Extract unique and distinguishable patterns related to particular age classes. That includes facial age features such as texture or edge relathioships of facial skin.

Global Features - apperance based feature extraction. The whole face is considered as feature space.

Local Features - geometry based feature extraction. Feature extraction from facial parts as eyebrow, nose, lips, etc.

2. Building classification or regression model

• Handcrafted-based approach

  based on combination of filters:

  • Histogram of Oriented Gradients (HOG) or Local Binary Pattern (LBP) to extract edges and shapes from facial image

• Deep learning approach

  based on CNNs. Examples of pre-trained models used for age estimation:

  • VGG-16, VGG-19, ResNet50, AlexNet, Xception, GoogleLeNet, MobileNetV2