Image style transfer based on local matching and global alignment
Yuhang Si, Dongzhi He · 2023
In recent years, image style transfer and deep learning technology have developed rapidly. It has been widely used in many fields, such as film and TV filters and art creation. However, the quality of the generated stylized images still needs improvement, often with content structure distortion, content detail loss, lack of semantic perception, and other problems. In this paper, we propose an image style transfer method based on local matching and global alignment. The local matching module is used to enhance the semantic similarity perception and improve the local style. The global alignment module improves the content structure distortion problem by aligning the statistical information of features. The integration of local matching information and global alignment information can better balance the style and content structure. The comparison experiments of stylized effects demonstrate that our approach can generate high-quality stylized images.