Revolutionizing Age and Gender Recognition: An Enhanced CNN Architecture

Abstract

The recognition of age and gender in images has had a significant impact on computer vision, particularly with the increasing application of digital platforms. Deep Convolutional Neural Networks (DCNNs) show promising performance; however, they demand substantial computational resources, limiting their deployment in real systems, especially those with constraints on resources or cost. This study performs a sensitivity analysis in order to show how some changes in the architecture of the network can influence the tradeoff between accuracy and performance. For that, in this work, we have investigated various CNN architectures and introduced an effective convolutional neural network (CNN) model to precisely predict gender and age attributes using the Adience dataset. Amidst unfiltered and diverse image sources from various devices, our model exhibits an impressive 92.24% accuracy across eight distinct age groups and two gender categories. The model's strength lies in its adeptness at handling intricate image data, allowing comprehensive adjustments to age and gender parameters. By employing advanced deep learning techniques and comparing with MiniVGGNet, our model showcases exceptional performance.