Art Image Style Conversion Based on Multi-Scale Feature Fusion Network

Abstract

To enhance the efficiency and quality of art image style conversion, this study improves the convolutional neural network style conversion algorithm by introducing a multi-scale feature fusion network, comprehensively considering different convolutional features, and combining attention mechanisms to extract important features of art images. It occupied less conversion time, CPU usage, and memory usage in the artistic image style conversion. It had better conversion performance. The research method had high peak signal-to-noise ratio and structural similarity index when converting different artistic styles. The highest peak signal-to-noise ratios for converting to Van Gogh art style, Ukiyo-e style, Monet style, and Cézanne style were 22.892, 17.844, 21.647, and 22.291, respectively, and the highest structural similarity index values were 0.842, 0.783, 0.845, and 0.843, respectively. The research has achieved effective conversion of target styles while preserving content in images, improving the quality and effectiveness of artistic image style conversion, and promoting the advancement of image processing technology.