pp. 3325-3332
S&M2692 Research Paper of Special Issue https://doi.org/10.18494/SAM.2021.3402 Published: September 30, 2021 Optimized and Improved Methods of Image Style Transfer for Local Reinforcement [PDF] Yong Li, Yan Wang, Hsien-Wei Tseng, Hongkun Huang, and Chun-Chi Chen (Received March 26, 2021; Accepted September 2, 2021) Keywords: image style transfer, deep learning, image segmentation, DeepLab2
Image style transfer, which commonly refers to adding a designated image style to a target content image, is now widely used in the movie industry, animation design, and game rendering, providing strong visual effects and cultural influences. However, there is no common criterion for evaluating the performance of image style transfer. In addition, people are more interested in local regions of images. This paper provides some revised methods to meet customer demand, focusing on an optimized image segmentation method based on DeepLab2, a semantic segmentation method, and fully connected conditional random fields (FCCRFs) for local image style transfer, with experiments demonstrating their usefulness and efficiency.
Corresponding author: Hsien-Wei Tseng, Chun-Chi ChenThis work is licensed under a Creative Commons Attribution 4.0 International License. Cite this article Yong Li, Yan Wang, Hsien-Wei Tseng, Hongkun Huang, and Chun-Chi Chen, Optimized and Improved Methods of Image Style Transfer for Local Reinforcement, Sens. Mater., Vol. 33, No. 9, 2021, p. 3325-3332. |