Analysing the Impact of Gender Classification on Age Estimation

Abstract

Age estimation from facial images is one of the most popular fields of research concerning deep learning and convolutional neural networks. However, there are several factors influencing the final accuracy that require special consideration, and in this research, we examine how gender classification affects age estimation. We use a predefined version of the MobileNetV2 convolutional neural network and train it on the CASIAWebFace dataset which we augmented with our private dataset called AgeCFBP. For the purpose of testing the network performance, we used the FG-NET dataset. The results of our experiments showed that gender pre-classification has a measurable impact on age estimation in both male and female population by decreasing Mean Absolute Error (MAE) metric, which might lead to enhanced applications in real-world scenarios, such as biometric authentication, security systems, human-computer interaction, and age-restricted content access control.