Research On Poetry Generate Picture Based On GAN

Ancient poetry is the treasure of our country's culture.The inheritance of poetry has become the cultural development requirement of the modern era.We often say that there are paintings in poems.There are poems in paintings.Poems and paintings not only give us visual enjoyment but also help us understand poetry better.At the same time,ink painting is a traditional Chinese painting and is also representative of traditional Chinese painting.The style of ink painting is more in line with the artistic conception of Chinese poetry,so it is possible to shift the style of ink painting to the resulting drawing.This article is precisely in this context,focusing on the two tasks of generating images and converting styles of images based on poetry.Through the research,we found that although there have been related researches on generating images based on texts,we have also achieved certain results.However,at present,the study of text-generating images is still limited to English.There are few studies on generating images based on Chinese texts.This paper takes ancient poetry as the research object.At present,there are no large-scale poetics and corresponding image data sets.The method of supervising is limited.Compared with modern texts,ancient poetry also brings challenges to the study of text-generating images.For the image style conversion task,most of the current image style conversion studies use a deep learning method.This method has a high training cost and the model is difficult to interpret.In view of the above research and analysis,this article will study the above two tasks respectively.(1)For the lack of current poetry-matching datasets,this paper builds a certain scale of correlations.For the data sets,we dealt with different ways of dealing with texts in ancient and modern Chinese.Experiments were performed on text-generated images on two different text features and scales of data sets.(2)In the work of poetry and painting,this paper consider that not all poetic vocabulary meanings are helpful for generating images.This paper classifies the subject matter of poetry based on the Support Vector Machine(SVM)combined with supervised methods.Image-independent semantic information.This paper builds a deep learning method based on Deep Convolutional Generation Adversarial Networks(DCGAN)and builds a model for generating poetic images.Compared with other methods,we have verified the effectiveness of the proposed method.(3)In the work of image style conversion,this paper considers the ink style black,white,and gray as the main colors.This paper uses Reinhard algorithm and wavelet transform method to construct the image style conversion model.We have verified that this method is simple,fast,and has good results in the overall style consistent(ink style)style transition.In general,based on deep convolutional generation adversarial networks can learn the approximate distribution of training data more,and the generated image is more realistic.We have also verified the effectiveness of the image style conversion method in this paper.