A System Of Generating 2D Animation From Video

With the advent of video segmentation, shape matching and non-photorealistic rendering techniques, novel approaches are proposed to allow video to be translated into a cartoon-like style. These approaches are different with the hand-painted celluloid one in traditional way, which is a branch of 2D computer animation.The paper describes a system for transforming an input video into a highly spa-tio-temporally coherent cartoon animation with a range of styles. The user simply need outline objects on keyframes, and the system automatically produces semantic regions for every frame. The user chooses a rendering style for the regions, and a kind style of cartoon is shown.The paper treats video as a space-time volume of image data, and has developed an anisotropic kernel mean shift technique to automatically segment the video data into contiguous volumes. To maintain smooth trajectories along the time dimension, a mean shift guided interpolation algorithm is employed to create spatio-temporal semantic regions by shape matching algorithm. At last, a sewing algorithm is used to sew the adjacent regions.