Style Transfer of Image Target Region Based on Deep Learning
Bingyang Niu, Yuxuan Ma, Yali Qi · 2021
With the rapid development of style transfer technology today, the global style transfer technology has basically taken shape, but in the actual application process, there are problems such as the inability to transfer the local style of the target area of the picture. Therefore, in response to such situations, this paper proposes a deep learning-based style transfer framework for image target regions. Firstly, the content map is the main input, and the mask map generated by the image mask technology is used as the specific condition input. Then, the transfer network style transfer area of the specific condition input is only the target area, and the non-target area is not in the style transfer area. Finally, the transfer network style transfer input via specific conditions realizes the image style transfer of the target area. At the same time, experiments were conducted on the public data set COCO 2014. The experimental results show that the target region style transfer network model based on deep learning proposed in this paper has a better local style transfer effect, and the transfer time has been improved.