Gender and Age Group Classification from Handwriting Using Lightweight Models on Edge Devices

Handwriting has unique characteristics for everyone, allowing for the classification of gender and age groups through analysis. This paper explores a lightweight model approach for classifying gender and age groups using Korean handwritten character images on edge devices. EfficientNet-Lite0 and MobileNetV3-Large models were converted into the tfLite format and applied to an Android application, where three-character word images were captured and classified using the device’s CPU and GPU. Experimental results showed that gender classification (two categories) achieved a maximum accuracy of 80.03%, while age group classification (five categories) reached 82.37%. When combining gender and age classification, the model achieved a maximum accuracy of 65.42%. This paper discusses the implementation and performance analysis of this lightweight model-based approach and suggests improvements for future performance enhancement.