Deep Learning-based Research on Stylistic Migration and Creative Assistance for Drawing Artworks

In recent years, deep learning algorithms have been gradually applied to the field of art creation, bringing new possibilities for art development. The study uses a generative adversarial network as the underlying logic of the image style migration model, and the improved CycleGAN method is used to assist in the style migration of sketching artworks to assist in their creation. After optimizing the CycleConsistent Generative Adversarial Network model, the loss function was designed to construct an improved GAN-based style migration model for sketch artworks. The CycleGAN model of this paper is compared with other image style migration models and retrograde algorithms in terms of loss, operation efficiency and image quality evaluation, so as to explore the performance of CycleGAN of this paper in sketch artwork style migration. Among all the image style migration algorithms, CycleGAN in this paper has the fastest convergence speed, the smallest number of parameters (20.75M), and the fastest running speed (3.42s, 2.19s, 1.72s). The CycleGAN model in this paper received the best subjective evaluation, with content quality, stylization strength, and favoritism exceeding 60%. The SSIM value and PSNR value of the CycleGAN model in this paper are larger than other models, and the optimal objective evaluation results are achieved.