Feature Guided Texture Synthesis (FGTS) for artistic style transfer

Abstract

Artistic images and paintings can be regarded as a composition of content and style. The aim of artistic style transfer is to synthesize a stylized image with content of source image but novel style from style example. Based on texture synthesis, a novel Feature Guided Texture Synthesis (FGTS) algorithm for artistic style transfer is proposed in this paper. Compared with existing example-based methods, the content of a source image is better defined in FGTS with a feature field generated from the source image. Though the style is modelled with low-level statistical features, the style transfer process is guided with the feature field which incorporates style and content during synthesis process. Moreover, a modified L neighborhood distance metric is developed to provide better measures of perceptual similarity. Results and comparisons are given to demonstrate that FGTS is an effective method for artistic style transfer.