Visual Analysis Of Song Ci Style Genre Discrimination And Image Style Transfer

Song Ci is numerous and has many different styles.With the continuous development and evolution,it gradually formed two styles,the wanyue school and the haofang school,which became the unique characteristics of Song Ci.Therefore,starting from the style genres and image style transfer of Song Ci,this paper analyzes the style schools of Song Ci from the perspectives of text,image and poets of Song Ci,so as to provide a new perspective of Song Ci study for lovers.Firstly,a multidimensional poetry data set oriented to Song Ci is constructed.Beautiful Soup method and Map method are used to process ancient poetry,and Song Ci text data set and poetry famous mountain data set are obtained.Combined with CBDB database,the social relation data set of Song Dynasty characters was obtained.In order to construct image style transfer data set oriented to Song Ci imagery,the open data set of Alice Xue and network pictures are used as the main sources.After pretreatment by Pillow method,the incomplete and abnormal images are artificially filtered,and the image sets of ink painting,wanyue style and haofang style are obtained.Secondly,considering the characteristics of Song Ci,which are concise and rich in meaning,in order to classify the wanyue school and the haofang school of Song Ci,a training corpus for determining the style of Song Ci is constructed by manual annotation.Word2 Vec method is used to convert the Song Ci training corpus into Song Ci word vector,and Text RCNN model combining Bi LSTM and CNN is introduced.According to the accuracy and loss value,the optimal model is selected to determine the style and genre of Song Ci.In order to verify the effectiveness of Text RCNN in the Song Ci style classification task,5 classic text classification models are selected to test the classification effect.The experimental results show that Text RCNN can obtain better accuracy.Then,in order to further show the different characteristics of Song Ci styles,the image style transfer method based on the Cycle GAN model is proposed to vividly show the different artistic conception of the wanyue school and the haofang school.Due to the distortion problem of image generated by the original model,Identity Loss function constraint generator is used to transform the image domain,L1 loss is used to calculate the similarity between the generated image and the original image,and the proportion coefficient of Identity Loss is dynamically adjusted to improve the image distortion and generate clearer stylized images.Finally,based on the visualization task,a series of charts such as multidimensional word cloud chart and six categories of people distribution chart are designed to depict the characteristics of word preference,common images and social relations of the two major styles of Song Ci,and the image style transfer method is used to show the characteristic differences of different schools,so as to conduct an in-depth exploration of the style genres of Song Ci.Using a questionnaire survey form for case analysis and visual analysis evaluation.The evaluation results show that the work is effective and practical.