Image Quality Improving in the Field of Style Transfer based on Deep Learning

Recent studies have made great progress in the transfer of style in the image and various methods have made progress in this field. However, evaluating and improving the quality of stylization remain two important open challenges. By examining these two aspects, in this paper, the style transfer quality is first decomposed into three measurable factors, namely Content Fidelity (CF), Global Effects (GE) and Local Patterns (LP). Then, a new approach is further presented to exploit these factors to improve the quality of stylization. The proposed approach is to use a deep neural network based on GAN or D-GAN in general, which uses the manifold model.

Image Quality Improving in the Field of Style Transfer based on Deep Learning Keywords:

Style Transfer , Image Quality , Deep Learning , D-GAN