Dynamic textures using wavelet analysis

Abstract

In this paper, we present a new dynamic texture technique by using our new mixed auto-regressive moving average exogenous (MARMAX) model in the wavelet domain. The technique captures much more detailed local and global dynamic texture properties than the previous works and it can synthesize long dynamic texture videos with temporal coherency. The whole input video is firstly transformed into the wavelet domain in different levels, then we use MARMAX model to capture the dynamic characters of the original video, finally we transform the video back by inverse discrete wavelet transform (IDWT). A desired long dynamic texture video can be synthesized easily from a short input one. The experimental results demonstrate that our approach can produce visually promising dynamic texture sequences. © Springer-Verlag Berlin Heidelberg 2006.