Description

Abstract

We proposed a new framework for dynamic texture segmentation based on spatiotemporal features. For the spatial mode, we employed a new texture feature to characterize each region of a frame of DT (Dynamic Texture), i.e., the histograms of LBP and WLD features in the XY plane of DT. For the temporal mode, we use the optical flow and the histograms of LBP and WLD features in XT and YT planes of DT to describe its motion field. We also addressed the problem of choosing thresholds for the segmentation framework. In addition, we developed a weighted Weber distance measure, which is computationally simple. Then results and comparison with existing methods show that our method is effective for DT (Dynamic Texture) segmentation. Furthermore, our method performs well on sequences with cluttered background.