A brief survey of dynamic texture description and recognition.pdfVIP

A brief survey of dynamic texture description and recognition Dmitry Chetverikov1 and Renaud Pe?teri2 1 Computer and Automation Institute, Budapest, Hungary csetverikov@sztaki.hu 2 Centre for Mathematics and Computer Science, Amsterdam, The Netherlands Renaud.PETERI@ 1 Introduction Dynamic, or temporal, texture is a spatially repetitive, time-varying visual pattern that forms an image sequence with certain temporal stationarity. In dynamic texture (DT), the notion of self-similarity central to conventional image texture is extended to the spatiotemporal domain. DTs are typically videos of processes, such as waves, smoke, fire, a flag blowing in the wind, a moving escalator, or a walking crowd. In physics, there is a long-established tradition of measuring, quantifying and visualising fluid and other flows that can be viewed as special kinds of temporal textures. In particular, Particle Image Velocimetry [12] is a standard technique for making a flow visible and measurable by injecting many small particles that scatter light and show the fluid motion. A frame of a PIV se- quence is a spatial texture; the whole sequence is a dynamic texture. Recently, there have been successful attempts to measure fluid flows using computer vi- sion methods such as optic flow estimation [6] and feature tracking [5]. The mutual influence of physics and image processing is obvious. For these reasons, it would seem natural for the vision community to learn from physics how to mathematically describe processes and motion patterns presented as dynamic textures. However, in computer vision tasks and approaches are quite different from those in physics. In pattern recognition, the study of temporal textures dates back to early nineties when the pioneering paper by Nelson and Polana [18] was published. Nelson and Polana categorised visual motion into three classes [24]: activities, motion events and temporal textures. Activities, such as walking or digging, are defined as motion patterns that