Research On Chinese Portrait Painting Style Transfer Algorithm Based On Deep Learning

With the burgeon of artificial intelligence,image style transfer technology has gradually become a research hotspot in computer vision.The style transfer and generation algorithm of portrait and other attributes is becoming the focal point,which has been widely used in human-computer interaction,face restoration,painting generation and so on.Chinese figure painting with unique oriental flavor is famous for its vivid use of pen to transport ink,in which portrait painting is the earliest branch.The creation of Chinese portrait painting not only requires the artist to accurately grasp the proportion distribution and facial features,but also needs to have profound Chinese painting creation skills,which is time-consuming and laborious.If we can quickly generate Chinese portrait painting with the help of style transfer technology,the efficiency of modern art creation will be greatly improved,which is of great theoretical and practical value.However,the current style transfer algorithm is mostly used in western painting,and a series of problems such as poor recognition,blurred,artifacts and even deformation often occur when directly used in Chinese painting.Therefore,on the basis of fully studying the cutting-edge style transfer technologies at home and abroad,this paper puts forward a style transfer algorithm suitable for Chinese portrait painting.Based on the task of deep learning style transfer towards Chinese portrait painting,the main contributions can be summarized as follows:1.A Chinese portrait painting style transfer(CPPST)algorithm based on CNN is proposed to convert a portrait photo into a Chinese portrait painting.Firstly,the brushstroke control restriction is proposed to guide the texture distribution of the image for freehand brushwork and fine brushwork of Chinese portrait painting.Then,the Chinese Painting Moving Distance is proposed to measure content and style features to harmonious transfer the style of Chinese painting to portrait photos.Finally,given the ink tone characteristics and the blank space reservation,we put forward the restriction for improving the loss network.The results show that the CPPST algorithm proposed can not only better show the identifiability and facial features of portraits,but also have a better Chinese painting style as a whole.2.A Chinese portrait painting generation(CPPG)algorithm base on GAN is proposed to solve the problem of poor robustness and limited input scene caused by face position and expression posture in portrait photos in the CPPST algorithm.Firstly,the CPPG algorithm uses the generation mode guided by the attention mechanism to resist the artistic style of network learning,and further migrates the face manuscript to the target domain of Chinese painting style,expanding the limited input scene.Secondly,aiming at the problem of inaccurate local color,strokes and other details of the generated results,CPPG algorithm adds the constraint of ink tone control reservation.The results show that CPPG algorithm can generate Chinese painting style face images with good visual effect.