Research On Chinese Character Generation Technology Basedon Generation Adversarial Network

Although the computer font library stores a variety of fonts such as Song Dynasty,Wei Style,and Huawen Caiyun,in the current era of personalization and the pursuit of personalization,people prefer to have a variety of personal style art fonts to meet their The growing spiritual demand for appreciation of the art of writing.Therefore,it is a necessary research direction to realize the generation of artistic style fonts.The main challenges faced by this problem are: 1)Compared with the Latin alphabet,the number of Chinese characters is large and the structure is more complicated.2)The mainstream Chinese character generation technology not only requires the construction of a large number of paired training sets,but also the generated Chinese character images have problems with unclear writing and distorted structures.3)Personal style fonts have greater style deviation and randomness.In recent years,deep learning models have been successfully used in various fields.In this paper,combined with the advantages of deep learning models such as strong data processing ability and high accuracy,two Chinese character generation models are proposed based on the generation confrontation model.It provides a potential solution to this problem.The main work of this article is as follows:1.In view of the scarcity of existing Chinese character pairing training data sets,and the font images generated by existing mainstream methods that have illegible writing and distorted structures,this paper proposes a multi-mode handwritten Chinese character generation model(HCC-GAN)based on a generation-resistant network.Specifically,by introducing self-attention,the network can pay attention to the texture geometric features of the text in the image,and improve the accuracy of the network model in generating Chinese characters.In addition,this paper optimizes the loss function,solves the problem that the existing method only calculates the similarity between the input font image and the generated image,but cannot reasonably measure the unaligned data,and implements two different domain font images.Similarity calculation process.Experiments on the CASIA-HWDB data set show that the model has a better Chinese character image generation effect.2.For the existing Chinese character generation method,it is only for one-to-one mutual conversion between two font fields of different styles,and cannot solve the problem of one-to-many mutual conversion between multiple style fonts.This paper proposes a multi-style handwritten Chinese character generation model(MSHCC-GAN)based on generative adversarial networks.Specifically,a new generator module(Gttention)is proposed,which can distinguish different styles while generating styles,so as to realize one-to-many conversion of the model to different style fonts.In addition,the loss function is optimized so that the model can maintain the consistency of the generated font image and the input font image in terms of content.The experimental results of the model on the CASIA-HWDB data set show that the MSHCC-GAN model can generate multiple fonts of different styles at the same time.