Deep Learning-based Painting Style Migration Algorithm and Its Visualization and Analysis

Computer graphics has made significant strides in the digital age, making image style transfer an exciting area of research. This study focuses on deep learning-based painting style transfer algorithms. By analysing classic style transfer algorithms such as GETIS and Johnson, we propose an improved algorithm based on attention mechanisms and multi-scale feature fusion. The results show that in terms of content-style balance, the improved algorithm based on the attention mechanism achieves a style fidelity score of 8 (compared to 7 for the Gatys algorithm), while the multi-scale fusion algorithm achieves a content retention score of 8. In terms of visual quality, the improved algorithm based on multi-scale fusion achieves an SSIM of 0.85, a PSNR of 33, and a MOS of 4.4, all of which outperform the classical algorithms.