Research On The Generation Of Handwritten Tangut Character Samples Based On Style Transfer

In the context of the current rapid development of artificial intelligence technology,the application of intelligent recognition technology to Tangut character recognition is bound to bring efficient scientific and technological research methods and resource organization methods for Tangut studies.However,due to the imbalanced distribution of the existing Tangut text samples,the training and recognition results of the deep learning model are largely affected.In order to solve the problem of incomplete sample categories due to the limitation of sample sources in the process of exploring the establishment of the Tangut text sample set,a study on the generation of handwritten Tangut text samples based on style transfer is proposed.Different from the existing style transfer applied to the generation of images,letters and Chinese characters,this article applies the existing style transfer technology to the generation of Tangut characters from standard fonts to fonts in ancient books.The main research content is as follows:(1)Research on the separation,extraction and reconstruction of the structural features and style features of the Tangut text image,to achieve the combination of structure and style,and recombine the given glyph structure and the desired style to generate the desired style of characters.(2)By analyzing the existing style transfer technology,designing and generating network models and identifying network models,the migration of Tangut characters from standard fonts to ancient book style fonts is realized.The generative network model studied includes text style feature and structure feature extraction and reconstruction modules,adopts encoding-decoding structure to design style feature extraction network,design structure feature extraction network based on U-NET network and residual network,and design identification network based on PatchGAN.The model verifies the validity of the generated samples through experiments.(3)The recognition accuracy of the original sample is compared with the recognition accuracy after the style transfer expansion.After the style transfer expansion,the accuracy of the Tangut text sample set on the depth recognition model has been greatly improved than the original sample recognition accuracy.Finally,the style transfer network model designed in this paper is used to generate Tangut character images,and a sample library of ancient books and documents handwritten style Tangut characters with complete sample categories and diverse sample appearances is constructed.The research on the generation of Tangut text samples based on style transfer proposed in this paper solves the defect of the unbalanced distribution of the original Tangut text sample data set,making the database can be used in various fields of Tangut studies and improving the research efficiency of Tangut studies researchers.