Dynamic Texture Analysis Using Eigenvectors of Gradient Skewness Tensors

In this paper, we propose a novel method for dynamic texture analysis based on the eigenvectors of higher-order tensors derived from the skew ness statistics of gradient values. We first introduce the gradient skew ness tensor for a video chip, and define its eigenvectors to characterize the properties of the dynamic texture. The eigenvector corresponding to the largest eigen value, which we use to describe the key dynamic patterns of the video, not only contains the illumination direction but also represents the changing nature of the movement over time. Considering these eigenvectors and their statistics as features, experimental results show that the proposed method is effective and robust for dynamic texture classification. Moreover, the eigenvector features can more subtly distinguish similar types of dynamic textures.