Feature Tracking

using Earth Mover’s Distance and Global Optimization

Abstract

Publication

Feature tracking plays an important role in the understanding of time-varying data sets since it allows scientists to focus on regions of interest and track their evolution and interaction over time. For complex data sets, previous feature tracking methods do not provide globally optimized matching results. Specically, there are two major drawbacks with the previous algorithms. First, the most similar attribute and the volume overlapping criteria used in these algorithms are inadequate in many cases. Second, previous algorithms are generally local tracking techniques, where individual features are matched independently with various greedy search schemes in the local surrounding.

These methods do not provide globally optimized match result in many cases. In this paper, we propose to use the Earth Mover's Distance (EMD) as the criterion to measure how well two features in adjacent time steps

correspond. EMD takes the full distribution of the features into consideration and has its advantage over the previous criteria. Furthermore, we use a global optimization algorithm to obtain the best match by nding the minimal EMD between the source and the destination feature sets. With the EMD criterion and the global optimization algorithm, features are tracked in a more accurate and efcient manner.

PDF  

Download