Chinese Font Generation Based On Generative Adversarial Networks And Chinese Character Prior Knowledge

Chinese character font generation technology is widely used in engineering practice and daily life,and has received a lot of attention from researchers.The current mainstream Chinese character font generation models are mainly based on deep learning,especially generative adversarial networks(GAN).Although existing GAN-based Chinese character generation models have achieved good results,there are still problems such as pattern collapse,especially in the task of calligraphic font generation.The main reason for pattern collapse is the lack of effective guidance information in the generation process of the model.To address these problems,this paper proposes two effective Chinese character font generation models that incorporate a priori information,such as the skeleton and contour of Chinese characters,into the font generation process.The main contributions of this paper are as follows:This paper addresses the pattern collapse problem in deep Chinese character font generation models by proposing an innovative approach called SGCE-Font.The proposed method incorporates the Chinese character skeleton into the existing deep Chinese character font generation model in the form of channel expansion.For Chinese character skeleton information,this paper uses a simple mathematical method to extract it effectively.Experimental results demonstrate the effectiveness of SGCEFont in alleviating the pattern collapse problem and outperforming existing models.Notably,the skeleton guidance module introduced in this paper can be applied to other existing deep Chinese character font generation models using plug-ins to further improve their generation performance.In calligraphic font generation tasks,the skeleton often struggles to fully capture pattern information,particularly when it comes to the calligraphic style.This limitation leads to suboptimal performance of the SGCE-Font model in calligraphic font generation tasks.To address this problem,we propose a novel calligraphic font generation model,named SIC-Font,which integrates the skeleton and contour a priori.The SIC-Font model incorporates the contour information of calligraphic characters into the SGCE-Font model and further guides the network to extract the global structural features of calligraphic characters with the help of Chinese character recognition technology by automatically constructing some non-exactly paired datasets.In the proposed SIC-Font model,we introduce a Canny operator to efficiently extract the contour information of calligraphic characters and an effective skeleton-contour fusion module to achieve efficient fusion of skeleton and contour.This method overcomes the limitations of previous models in characterizing calligraphic style information and improves the performance of the model on the calligraphic font generation task.The experimental results demonstrate that the introduced contour information and the automatic matching dataset based on Chinese character recognition provide more useful information for calligraphic characters,resulting in better generation results than existing deep calligraphic font generation models.