Handwritten Ancient Chinese Character Recognition Algorithm Based on Improved Inception-ResNet and Attention Mechanism

Abstract

Chinese characters have distilled the Chinese nation's vast wisdom and values, but the general public's learning and enjoyment of ancient scripts are hampered by the fact that fonts from different dynasties have highly different styles, intricate structures, and diverse deformations. To solve the difficulty of ordinary people identifying ancient Chinese characters, an ancient font recognition system that is based on an improved Inception-ResNet network is proposed. ECA-Net is integrated into the Inception module, PReLU is utilized to activate the network, and the Nadam algorithm is used to enhance the model training effect. The experimental results demonstrate that the method outperforms the other five deep learning models with a recognition rate of 95.56% in the mixed font dataset consisting of six types of fonts: seal script, inscription, running script, cursive script, clerical script, and regular script.