Unsupervised Dynamic Texture Segmentation Using Appearance and Motion

Dynamic texture (DT) is an extension of texture to the temporal domain. How to segment DTs is a challenging problem. In this paper, we address the problem of segmenting DT into disjoint volumes in an unsupervised way. DTs might be different from their spatial mode (i.e., appearance) and/or temporal mode (i.e., motion field). To this end, we develop a framework based on the appearance and motion modes. For the appearance mode, we use a new local spatial texture descriptor to describe the spatial mode of DT; for the motion mode, we use the optical flow and the local temporal texture descriptor to represent the temporal variations of DT. In addition, for the optical flow, we use the Histogram of Oriented Optical Flow (HOOF) to organize them. To compute the distance between two HOOFs, we develop a simple, effective and efficient distance measure based on Weber Law. Each volume is characterized by its appearance and motion modes. Furthermore, we also address the important problem of threshold selection by proposing a method for determining thresholds for the segmentation method by statistical learning. Experimental results show that our method provides very good segmentation results compared to the state-of-the-art methods in segmenting volumes that differ in their dynamics.