A local descriptor for finding corresponding points in vector fields

Vector fields may come from video data (via optical flow and tracking), from weather phenomena (e.g., wind speed and direction), and from medical imaging. An important component in analyzing this data is to be able to quantitatively compare different points within a vector field or across different fields of the same type. We present a novel local descriptor to compare individual points in a vector field and rank them based on which are most similar to a selected point. The descriptor captures the statistics of the neighboring vector field around a central point and is discriminating, robust to noise, and efficient to generate and compare.