Extraction and Reconstruction of Traditional Art Visual Elements by Graphic Design Incorporating Deep Learning

Abstract

Abstract In today’s digital trend, deep learning technology has become an indispensable innovation power in the field of graphic design. In order to better integrate traditional visual elements from graphic design into graphic design, this paper proposes a graphic design layout generation model based on the LayoutGAN model. Taking Chinese traditional paper-cutting art as an example, we construct a dataset of visual elements and use Gaussian filtering, the LGE model, and the OTSU algorithm to preprocess the images of visual elements in paper-cutting art. The LayoutGAN model utilizes the layout element wireframe rendering module to enhance the extraction of visual elements of traditional art and a 2-stage training strategy is employed for model optimization training. To demonstrate the feasibility of applying the model in the reconstruction of visual elements of graphic design art, it was tested experimentally. After the LGE model was used for the image enhancement of the visual elements of paper-cutting art, the image enhancement effect was improved by 30.68%, and the segmentation accuracy and the mIoU of the traditional art visual elements were 98.24% and 97.58%, respectively. The PSNR value of the graphic design generated using the LayoutGAN model is 21.27, and the subjective evaluation scores are all above 3. The reconstruction of traditional visual elements in graphic design can be achieved using deep learning models, and the original information is preserved to the maximum extent.