Vector Field Segmentation Based on Integral Curve Attributes

We propose a segmentation method for vector fields. Our segmentation is driven by integral curve attributes that are used to classify different behaviors of integral curves. In particular, we assign an attribute value to each spatio-temporal position based on the integral curve that passes through it. With this attribute information, our segmentation first performs a region classification. Then, the connected components are constructed from the derived classification to obtain an initial segmentation. After merging and filtering small segments, we extract and refine the boundaries of the segments. Because the points that are correlated by the same integral curves have the same or similar attribute values, the proposed segmentation typically results in segments that are well-aligned with the flow direction. Therefore, additional processing is not required to generate other geometric descriptors within the individual segments to illustrate the flow behaviors. We apply our method to a number of synthetic and CFD simulation data sets and compare it with existing methods to demonstrate its effectiveness.