Research And Implementation Of Color Ink Painting Generation Based On Style Transfer Technology

Color ink painting is a form of traditional Chinese painting,characterized by the refinement and simplicity of brushstrokes,ink colors,and forms on a complex line drawing basis.It creates aesthetic effects with intrinsic artistic value and is an indispensable part of Chinese painting art.At the same time,color ink painting also requires certain skills,and the painter needs to have a certain foundation and painting skills.With the development of technology,the application of computer vision technology has become increasingly widespread.Image style transfer technology,as an important technical means,provides new ideas and methods for the generation and artistic creation of color ink painting.The image style transfer is a technique that can change the style of an image while preserving its content,commonly used for image style conversion and enhancement.However,traditional image style transfer techniques have significant shortcomings,including poor generation results,high computational complexity,and difficulty in control.Cycle GAN,as an unsupervised image style transfer method based on deep learning,effectively solves these problems by introducing adversarial loss and cycle consistency loss,achieving good results.The article focuses on the task of generating color ink paintings by applying unsupervised style transfer methods based on deep learning to Chinese color ink paintings.To achieve this goal,the authors constructed a dataset containing both real landscape photos and color ink landscape paintings,aiming to enable the model to learn the typical features of color ink paintings and generate realistic images.However,due to the generalization of Cycle GAN,which treats different styles of transformation equally,it has certain limitations in generating color ink paintings.Therefore,the article proposes a series of improvement methods to address different issues.Firstly,this paper introduces the attention mechanism and extracts the brightness map of the image based on the idea of self-attention mechanism,reducing the weight of the sky part of the input image.Then,an attention module CBAM is added to the encoding part of the generator to guide the network to focus on important areas,improving the clarity of the generated images.Secondly,the Ada LIN function is introduced in the decoding part of the generator to enhance the network’s style feature learning ability and make the generated images more stylistically similar to Chinese ink paintings.Finally,in order to further improve the quality and aesthetics of the generated images,this paper introduces the structure loss and color loss based on edge detection algorithm and multi-scale structural similarity,respectively,which preserve the detail information and color characteristics of the original image,improving the realism and color vibrancy of the generated images.This paper confirms the effectiveness and superiority of the final model through a series of comparative experiments,ablation analysis,and quantitative analysis.Meanwhile,user surveys indicate that this paper’s method can generate high-quality color ink paintings that are more realistic,natural,and popular among the public.