Extended Statistical Landscape Features For Dynamic Texture Recognition

This paper concerns on the description and recognition of dynamic textures. The analysis of dynamic textures is an extremely important problem in the field of computer image processing and computer vision. The description and recognition is a key research issue of dynamic textures. Dynamic textures, which are the sequences of video images, not only contain the static information derived from the structure of object, but also contain the rich motion information of object. Nowadays, object recognition concerns mostly on the static features such as: geometric, photometric. The extracted information is not enough. Analyzing the motion process of object, extracting the dynamic features of object motion and integrating the static features can promote the recognition capability to the special object.Dynamic textures exist extensively in the nature, such as the motion of smoke, stream of water, ocean wave, a flock of birds hovering, etc. It is due to the dissimilarities between objects motion in the three dimensional space and the observation plane in the two dimensional space, and the change of illumination while object moving, and the cover of other object, and the quasi periodicity of many objects motion that make the recognition of dynamic textures become the most challenge study.This paper carried on the research to existing method of dynamic texturesanalysis, and proposed the method of Extended Statistical Landscape Featuresto describe and recognize the dynamic textures. The main work is as following:1. The static textures and the description method: we first introduce theconcept of static textures, and the main existing methods of descriptionand recognition for static textures. The method of Statistical LandscapeFeatures, which was used in our experiment, is also discussed in detail.Experiment results show that it achieves a higher recognition performance for the static textures.2. Dynamic textures and Optical Flow: we introduce the concept and classify method of dynamic textures, the concept of Optical Flow, the constraint equation of optical and the computing method of Optical Flow. Optical Flow hammers at the building of optical flow field of object motion. Due to the two dimensional space of observation plane, optical flow can not describe the motion in three dimensional space rightly. However, Optical Flow still is the main method in dynamic textures analysis now.3. Dynamic texture description based on Extended Statistical Landscape Features: after two hypotheses in Optical Flow which are Brightness Constancy Model and Gradient Constancy Model were analyzed, we improved on Block-matching Optical Flow calculation and proposed Local Motion Pattern which was used to describe the dynamic features. Extended Statistical Landscape Features uses Statistical Landscape Features to describe static features and Local Motion Pattern to describe motion features. The histograms were used to recognize dynamic textures, which were concatenated by the Local Motion Pattern histogram and the Statistical Landscape Features histogram. At last, the latest study progress is also showed in the paper.4. Experiment evaluations, which are based on DynTex dynamic texturedatabase, are presented at the last part.