Research On The Method Of Chinese Ink-Wash Painting Generation Based On Samples

Chinese painting,having a long history,is one of the important cultural heritages in China.However,it is difficult to draw a satisfactory Chinese painting since it usually requires years of professional training.Ink-wash painting is the representative of Chinese painting and how to create images with ink-wash style is one of the challenging topics in computer science.Combining computer technology with traditional Chinese culture to automatically generate images with ink-wash features based on the sample images is not only helpful to fill in the deficiencies of traditional neural style transfer algorithm in transferring ink-wash painting,but also to inherit national culture,which has practical significance and research value.This thesis studies the generation of ink-wash painting and proposes an improved neural style transfer algorithm based on the sample images,i.e.,given a sample image,and transfer the style features in the sample image to the content image to generate an ink-wash painting image with the same style.This algorithm considers the unique feature of ink-wash painting,including ink diffusion,yellowing,and space reservation,so that the generated images with ink-wash style are more realistic.In summary,we make the following researches and contributions: 1.We summary the unique feature of the sample ink-wash painting.Then we study the algorithm principles and design implementation of the traditional neural network style transfer techniques.The analysis points out the problems and solutions for improving the performance of traditional neural style transfer algorithms.And we use Holistically-Nested Edge Detection algorithm to obtain the edge prediction results of the content image at multiple scales and levels.Combined with the features of ink-wash painting,an edge fusion method for obtaining edge lines and an abstract fusion method for obtaining the abstract skeleton of the image are proposed.2.In order to solve the problem that traditional neural style transfer algorithms cannot well express the ink and wash diffusion feature of ink-wash painting,we perform Gaussian blurring and image dilation on the brush-fusion image,so that the edge lines have a blurring effect similar to the ink diffusion,thereby simulating the ink diffusion effect.3.In order to solve the problem that traditional neural style transfer algorithms cannot express the yellowing feature of ink-wash painting,the yellowing loss is adopted to make the resulting image consistent with the ink style image in hue,so as to simulate the yellowing effect.4.In order to solve the problem that traditional neural style transfer algorithms cannot express the space reservation features of ink-wash painting,we perform yellowing processing on the obtained abstract fusion image,and then appropriately fuses with the intermediate result image during iterative training to deepen or weaken part of the image area respectively to simulate space reservation effect.5.We conduct extensive experiments on various types of sample ink-wash painting,then comparative analysis is performed from three aspects,including visual quality,user research,and ablation experiments to verify the effectiveness of the improved algorithm in this paper.The experimental results show that compared with the traditional style transfer algorithm,the proposed method in this paper can better express the feature of ink and wash diffusion,yellowing,and space reservation,and the results are closer to the sample ink-wash painting style.