CoAtNet-HOG Feature Fusion for Efficient Font Style Classification

Font style plays a critical role in the standardization assessment of Chinese characters. Non-standard fonts may lead to blurred strokes and structural imbalance, compromising evaluation accuracy. To address this, we propose ML-HCFP, a novel model integrating CoAtNet with HOG features that achieves 99% classification accuracy. Furthermore, through knowledge distillation, we develop the lightweight DHCFP-ResNet18 variant which maintains 98% accuracy while reducing parameters by 103.8MB. This research significantly enhances font recognition precision and efficiency, providing robust technical support for Chinese character standardization assessment.