Generating Video Textures by PPCA and Gaussian Process Dynamical Model

摘要

Video texture is a new type of medium which can provide a continuous, infinitely varying stream of video images from a recorded video clip. It can be synthesized by rearranging the order of frames based on the similarities between all pairs of frames. In this paper, we propose a new method for generating video textures by implementing probabilistic principal components analysis (PPCA) and Gaussian Process Dynamical model (GPDM). Compared to the original video texture technique, video texture synthesized by PPCA and GPDM has the following advantages: it might generate new video frames that have never existed in the input video clip before; the problem of "dead-end" is totally avoided; it could also provide video textures that are more robust to noise.