Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Electrical and Electronic Engineering

In the realm of machine learning, estimating age from photographs of individuals is a
difficult challenge. It has gotten a lot of interest from academics in recent years because
of its uses in fields including customized advertising, content access regulation, and
surveillance systems. Building an effective age estimation network, on the other hand,
presents various obstacles, including data discrepancy, the quality of facial images,
and individual aging trends. This study provides an in-depth examination of the typical
approaches for developing age estimate models with the use of Convolutional Neural
Networks. It also addresses standard datasets for training and assessment, as well as
the most recent advances in this field. The research also looks at the common
assessment measures that are used to evaluate the effectiveness of age estimation
algorithms. This study's main contributions are the development of an effective
Convolutional neural network age estimation model with quality performance and a
thorough analysis of the impact of various factors on the model's performance, such as
data preprocessing, data augmentation, network architecture (substituting some
convolution layer with separable convolution to help fight overfitting due to small
dataset), and transfer learning. The report not only provides a survey but also
highlights gaps in the currently available information and makes recommendations for
further research.