AgeTrans: Estimating the Age of Faces Using Global Features

To meet many different needs in practice, the estimation of the age via face images is of great interest and this is also one of the important tasks of the machine vision field. Therefore, there are currently many methods that can solve this task, typically methods using Convolution Neural Network (CNN). However, the structures using CNN method can only extract local features of the face and not showing much of the relationship between facial parts. Therefore, we propose a solution called AgeTrans that combines Transformer method and CNN method to create an “end to end” model with better efficiency. Specifically, taking advantage of the Attention mechanism of Transformer method to extract the global feature of the face and using CNN method to extract the local feature, then logically combining the two types of features and giving the result of the task. From the experimental results, our model is more efficient and competitive enough on the UTKFace dataset and at the same time, confirms that it is possible to use global features to estimate the age of the face.