Dynamic Texture Classification using Combined Co-Occurrence Matrices of Optical Flow

This paper presents a new approach to Dynamic Texture (DT) classification based on the spatiotemporal analysis of the motion. The Grey Level Co-occurrence Matrix (GLCM) is modified to analyse the distribution of the magnitude and the orientation of the Optical Flow which describes the motion. Our method is therefore called Combined Co-occurrence Matrix of Optical Flow (CCMOF). The potential of a multiresolution analysis of the motion is revealed by experimentation. We also demonstrate the importance of the analysis of motion in the spatiotemporal domain. Finally, we demonstrate that adding a spatiotemporal motion analysis (CCMOF) to an appearance analysis (Local Binary Patterns on Three Orthogonal Planes (LBP-TOP)) significantly improves the classification results.