Research On The Calligraphy Image Generation Base On Generative Adversarial Networks

Calligraphy is a unique writing carrier of Chinese characters.From Oracle,gold and official script,calligraphy has gradually developed into regular script,cursive script,calligraphy and so on.Nowadays,it is very difficult to preserve the paper and inscriptions as the carrier of calligraphy works inheritance.The natural factors such as oxidation,moth,wind erosion and war disorder and damage have a great influence on the preservation of calligraphy works.Therefore,it is very meaningful to use the present science and technology to repair the ancient calligraphy books and inscriptions in China.In the field of computer vision,convolutional neural network is used to classify and extract features,and the research on image restoration with the generated neural network has been very good.These theoretical results provide a better way to repair,preserve,promote and develop calligraphy culture.The existing image restoration model usually uses the picture overall information to repair the local information.Because of its particularity,the whole picture consists of many characters in a specific location.There is a certain relationship between the position and semantics between the characters,but the relationship between the two characters is not very big.When the broken degree of calligraphy picture is large,the existing image restoration method can not infer the word of the damaged position according to the remaining pictures to repair the whole picture.In view of the existing technology,this paper proposes a model of generating calligraphy pictures based on the generation of counter network,and generates corresponding calligraphy pictures according to the text.The main work of this paper is as follows:1?The position prediction of calligraphy characters.Generating a calligraphy picture is different from generating a single calligraphy word.This task requires knowing the position of each calligraphy word in the picture.The paper proposes a calligraphy location prediction model to predict the relative position of a text in the picture.2?Combine calligraphy words and position to generate pictures.This paper uses the word coding vector to generate the picture of calligraphy words.Combined with the location prediction generated for each word,it is integrated into a multi-channel feature picture,and the multi-channel image is input into the high-definition generated network to complete the generation of calligraphy words.3?The model of the formation of various calligraphy characters.The position prediction of calligraphy characters in the model proposed in this paper mainly comes from the vector of text editing,so it has no good generalization ability.In the improved method,the feature extracted from the calligraphy image is used to predict the position coordinate of the calligraphy character.The improved model used neural network to analyze and extract the characters of the strokes and strokes,and finally used to predict the position of calligraphy characters.The trained model can directly analyze the shape of the cut rectangle of the word image prediction word during the use process,so that the model does not need to have corresponding processing method for each word.