Optimizing Image Quality for CNN-based Image Style Transfer

Yi-Hsuan Hung

doi:https://doi.org/10.36463/idw.2023.0737

13:50 〜 14:10

[VHF3-2] Optimizing Image Quality for CNN-based Image Style Transfer

*Yi-Hsuan Hung¹, Sn-Jin Chen¹, Pei-Li Sun¹, Zhon-Yu Chen², Ming-Jin Huang², Chia-Pin Chueh² (1. National Taiwan University of Science and Technology (Taiwan), 2. Printing Technology Research Institute (Taiwan))

image style transfer, super-resolution, AI, image quality

https://doi.org/10.36463/idw.2023.0737

Proposing a method to improve naturalness and clarity of images generated by CNN-based image style transfer. It includes analyzing the content image and the style reference to optimize the content-weight, using Super-resolution model to enhance the sharpness, and giving face regions higher content-weights to improve naturalness of the generated images.

Proceedings of the International Display Workshops Volume 30 (IDW '23)

[VHF3-2] Optimizing Image Quality for CNN-based Image Style Transfer