Application and optimization of image style transfer in modern graphic design

The rich micro level and fine-grained features in digital images are often presented in a mixed manner, making it difficult to effectively separate and preserve them in style transfer, which increases the difficulty of technical implementation. To enhance the artistic expression of graphic design, this article studies the application and optimization methods of image style transfer in modern graphic design. Firstly, an unsupervised image style transfer model is constructed, which takes the original image and its contours as dual inputs, extracts style features through two encoders, and inputs them into a generator to generate stylized results. Then, two discriminators are introduced to perform multidimensional feature discrimination on the generated results. To effectively preserve the micro level features in the image, this study introduces stroke constraint and contour constraint mechanisms to guide the discriminator to focus on the style characteristics conveyed by the strokes and the style attributes carried by the contours, respectively, improving the delicacy of style analysis and the visual consistency of output results. Finally, the optimized style transfer image is imported into graphic design software for layout adjustment and detail optimization. The experiment shows that this method has strong ability in image smoothing processing and can accurately extract image style feature vectors. The visual effect of the style transfer image obtained after stroke and contour optimization is better, and it can also be effectively applied to poster design, with significant application effects.