Dynamic Texture Classification Based on Motion Statistical Feature Matrix

Abstract

Dynamic texture is an extension of texture to the spatio-temporal domain. The motion property is one of the most important characteristics in dynamic texture analysis. This paper proposes a dynamic texture classification method based on the block-based motion estimation method and the statistical feature matrix. The method makes use of the block-based motion estimation method to estimate the motion vector of the dynamic texture, and constructs the statistical feature matrix by computing the complex covariance and dissimilarity of the motion vectors, which is used to describe the dynamic texture. The classification method is evaluated on two benchmark dynamic texture databases. The experimental results demonstrate the effectiveness of the introduced method.