Research On The Model Of Mongolian Font Style Transfer Based On CNN And CGAN

The creation of Mongolian fonts is a relatively complex and difficult task.Different from the specific writing structure of Chinese characters,Mongolian is a pinyin text with the writing from top to bottom and left to right.It consists of 35 characters.The font is based on the main body and is divided into three forms: the beginning of word,the middle of word and the end of word.Its special feature is that the shape of the same element at different positions of the word also change accordingly;the distance of writing between the letters follows certain rules,otherwise it is very easy to cause confusion;according to the pronunciation of the letters,the fonts will change in different positions of the word.Due to the large number of Mongolian characters and complex fonts,designing a set of Mongolian fonts with a specific style requires a large number of professional researchers who are proficient in the Mongolian language field to rely on computer assistance or use traditional manual methods such as rubbing and knife cutting to analyze each Mongolian font.Each part is individually designed.For such tedious work,it takes a lot of time,and it may even take several years to design a set of Mongolian fonts with a unique style,which consumes a lot of time,manpower,and material resources.The above-mentioned problems have caused designers to strive for a faster way to design Mongolian fonts.Therefore,a model that can automatically generate Mongolian font styles is very necessary.At the same time,non-Mongolian professionals can also design their own Mongolian font style.With the increasing development of deep learning,scholars at home and abroad have carried out research on the automatic transfer of Chinese characters and English font styles,but the Mongolian field is still in a blank stage.Therefore,this thesis attempts to use the deep learning methods to study the Mongolian font style transfer task,and implements the corresponding algorithm.The specific work content is as follows:(1)In view of the current situation of Mongolian font design,the background and significance of the research tasks of this thesis are explained,and the current domestic and foreign research status in the field of deep learning and style transfer is investigated and summarized.At the same time,the content of this article is repeated.(2)The thesis first uses TTF or OTF files to create a Mongolian font data set,and performs normalization preprocessing on it.(3)Constructing a Mongolian font style transfer model based on convolutional neural network.This thesis first uses convolutional neural network for experiments,and builds a convolutional neural network model structure.The model uses root-mean-square optimizer to automatically adjust the learning rate,reduce the difference value gradually and generate Mongolian from Mongolian title fonts to handwriting fonts directly.The generated font style is basically close to the real font style,achieving the effect of font style transfer.(4)Constructing a Mongolian font style transfer model based on a conditional generation confrontation network.The model uses generation loss and discriminant loss to measure the model.The Adam optimizer automatically adjusts the learning rate and gradually reduces the difference value until the generator and discriminator reach the Nash equilibrium state during the confrontation between each other.It can directly generate Mongolian handwritten fonts from Mongolian title fonts to achieve a better Mongolian font style migration effect.