A spatiotemporal context phrase description for general dynamic texture

In this paper, we propose a novel dynamic texture description method base on spatiotemporal context phrase for general dynamic texture. Different with the existing methods, we consider the spatiotemporal context both in the feature extraction phase and in the feature description phase. We present a space time constraint and salience rank strategies to extract the representative interest points. Then, we propose a novel space time context phrase method to mining and describe the semantic and spatiotemporal correlation of interest points. Finally, the space time context phrase is used in the nearest neighbor classifier to classify dynamic texture scene. We test our algorithm on the dynamic texture classification and human action classification tasks on the Dyntex dataset and the KTH dataset, respectively. The results show that our proposed method outperforms the state-of-the-art methods on the tasks.