Advancing Bangla typography: machine learning and transfer learning based font detection and classification approach using the ‘Bang-laFont45’ dataset

This paper presents a dataset for detecting and classifying Bangla fonts, consisting of 28,000 images across 45 classes, aimed at supporting font users and typography researchers. Four traditional machine learning models— Support Vector Classifier (SVC), Logistic Regression (LR), K-Nearest Neighbors (KNN), and Random Forest—achieved accuracies of 93.43%, 92.37%, 84.71%, and 81.48%, respectively, with SVC performing best. Six transfer learning models—VGG-16, VGG-19, ResNet-50, MobileNet-v3, Xception, and Inception—were trained, yielding accuracies of 87.74%, 80.00%, 87.26%, 80.55%, 82.30%, and 80.11%, respectively. The results highlight the effectiveness of both traditional and transfer learning models in font detection, with SVC and VGG-16 emerging as top performers.