Artistic Style Learning

This paper presents a novel example-based non-realistic rendering approach. In our algorithm, two basic texture synthesis approaches are used to find matching pixels in the source images: an approximate search, which is used to find the closest-matching pixel; and a coherence search, based on Ashikhmin's approach, which attempts to preserve coherence with the neighboring synthesized pixels. Painting styles can be transferred automatically from source images to the target image, and different effect results can be simply adjusted with several parameters. Furthermore, a new searching method based on KNN is also proposed to speed up texture synthesis. Experimental results demonstrate our algorithm is very efficient for style transfer.