Shape context based video texture synthesis from still images

A video texture provides a continuous varying stream of images. In this paper, we introduce a new algorithm that allows the user to easily produce a visually plausible video texture from a small collection of still images. In order to approximate the original temporal order of the scene, we apply a shape context based order recovery algorithm to the still images. The frame sequence is subsequently generated using a second order Markov Chain model. The final video texture is generated by frame interpolation which utilizes thin-plate spline warping and inverse distance weighting interpolation techniques. In conclusion, we show that significant improvements are obtained by using TPS warping technique instead of traditional optical flow method.