Font Size: a A A

Research On Feature Modeling And Style Transfer Technology Of Calligraphy Works Based On Deep Learning

Posted on:2024-01-23Degree:MasterType:Thesis
Country:ChinaCandidate:S M GengFull Text:PDF
GTID:2545306920463394Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Chinese calligraphy is an ancient art of calligraphy with a history of thousands of years and a unique charm.Chinese characters,as the official standard characters used today,contain unique meanings.Therefore,it is of great significance to use modern computer technology to transfer the content and style of calligraphy.With the rapid development of deep learning technology,it has been applied to various aspects such as image recognition,which has greatly improved the shortcomings of traditional methods in recognition efficiency and accuracy.This paper proposes a method of calligraphy recognition and style transfer based on deep learning.Convolutional neural network is applied to calligraphy recognition,and works with specific calligraphy features can be generated by using adversarial generative network and style transfer.The main contents of this paper are as follows:(1)After building the calligraphy data set network based on ResNet50 model and VGG16 model,an improved VGG16 network is proposed.Introduce a custom residual block.The continuous convolution layers in the VGG16 model are regarded as a convolution group,and an additional branch is added at the input of each model of the five convolution groups.The branch directly connects the input to the back of each convolution group,and the convolution results are added to the input data to form new features.According to the size of the two branches of the data,the different categories are added.The recognition accuracy of the improved model is significantly improved.(2)In the research of calligraphy style transfer technology,an improved Pix2pix network is proposed.In order to solve the problem of poor tolerance and robustness of traditional GAN network,spectral normalization and residual module were introduced into Pix2pix network.In order to improve the details and overall effect of the generated calligraphy images,the perception loss and attention mechanism are introduced.The improved model has better results.(3)Based on the network model,the interface display system is designed and developed.Through the interface to interact with the user,on the basis of collecting and processing the input data sources needed by the network,the results of the multidimensional analysis model operation are obtained,the potential information is mined,and the identified types of calligraphy content or generated diverse calligraphy styles are displayed to the user.
Keywords/Search Tags:Convolutional neural network, Calligraphy recognition, Transfer learning, Adversarial generative network
PDF Full Text Request
Related items