Streamline similarity analysis using bag-of-features

Streamline similarity comparison has become an active research topic recently. We present a novel streamline similarity comparison method inspired by the bag-of-features idea from computer vision. Our approach computes a feature vector, spatially sensitive bag-of-features, for each streamline as its signature. This feature vector not only encodes the statistical distribution of combined features (e.g., curvature and torsion), it also contains the information on the spatial relationship among different features. This allows us to measure the similarity between two streamlines in an efficient and accurate way: the similarity between two streamlines is defined as the weighted Manhattan distance between their feature vectors. Compared with previous distribution based streamline similarity metrics, our method is easier to understand and implement, yet producing even better results. We demonstrate the utility of our approach by considering two common tasks in flow field exploration: streamline similarity query and streamline clustering.