Patch-based Painting Style Transfer
Fay Huang, Chia-Lin Chien · 2020
Approaches based on convolutional neural network framework have proven to be effective in various of style transfer scenario. Most approaches train the network to process the photo as a whole, and gradually transform it into a user specified painting style. In contrast to that, this paper proposed a brush-based concept, which inherits the spirit of the stroke-based rendering. The input photo was first partitioned into small patches. A modified version of the generative adversarial network has been adopted to transform each patch into a brushstroke of a given style. After merging all the generated brushstrokes, it produces the desired style transfer result. The proposed style transfer approach preserves the geometrical structure of the input image; only color and texture are to be altered during the process.