Video Painterly Animation System Based On Layered Dense Correspondences

Image and video stylization is one of major topics in Non-PhotorealisticRendering (NPR) in computer graphics. Image stylization produces stylizedimage by using information from images. One the other hand, videostylization produces stylized video by analyzing information from videos,propagating effects and taking advantage of image stylization techniques.Image and video stylization is widely used in animation and entertainment,and gains increasing attentions from researchers. In recent years, imageand video stylization achieves impressive results upon the greatprogresses in computer graphics.The main components of stroke based video stylization rendering areintra-frames correspondences extraction, frame stylization and strokepropagation. Frame stylization determines the final style of output video.Stroke propagation propagates strokes to the rest of frames by applyingdense correspondences. The accuracy of correspondences determines thedegree of visual flickering and artifact in final outputs.In this paper, we propose layered adaptive optical flow (LAOF) bycombining optical flow and motion segmentation for dense correspondencesextraction. Our method optimizes dense correspondences in an iterationmanner. Firstly, our method uses adaptive optical flow to extract densecorrespondences among frames. Then our method clusters the motion vectors into motion layers and takes layers information as new parameters toadaptive optical flow. Besides, our framework uses beforehand renderingto alleviate visual flickering caused by stroke propagation.In this paper, we summarize a general4layer framework, and implementour video stylization system based on it. Our system provides layereddense correspondences extraction and stroke propagation technique, canwork with any kind of stroke-based rendering (SBR) techniques to producestylized videos.In the experiments, we test our algorithm by experiments. After that,we apply our method to videos with different kinds of motion. It de-monstrates the advantage of our method in alleviating visual flickeringin occlusion and non-rigid motion areas.