Deep Learning-Based Age and Gender Classification for Accurate Recognition

Abstract

Because of its widespread use in facial analysis tasks, automatic gender and age prediction from facial photos has attracted a lot of attention. Because of the wide intra-class diversity in face images—including variations in lighting, position, size, and occlusion—current models frequently fail to achieve accuracy. These problems make it hard to use them in practical situations. To accurately predict gender and age group, we present a deep learning architecture that merges attentional and residual convolutional networks in this study. Our model is able to improve prediction accuracy by zeroing in on important and relevant areas of the face thanks to the attention mechanism. We use a multi-task learning strategy to enhance the accuracy of age prediction by adding the projected gender to the age classifier's feature embeddings. We have trained our model using a popular dataset that includes information about the age and gender of faces, and the results are encouraging. We also demonstrate that the trained model has learnt to focus on the important facial features—the eyes and the mouth, for example—by visualizing its attention maps; these features are critical for precise age and gender categorization. For practical uses involving face recognition, this method offers a solid answer.