Using optimized gaussian mixture model rules and global tracking graph for feature extraction and tracking in time-varying data

Advances in computational power and numerical simulations have made many complex scientific simulations possible, generating large-scale and complex time-varying data. How we effectively extract and track the features contained in time-varying data play a crucial role in helping scientists recognize and understand the trendsD and dynamic behaviors behind these simulations. Many feature extraction and tracking methods often require the user to initially feed numerous feature data, e.g., a volume into their models for feature extraction, and then track the feature locally by comparing the features at two consecutive time steps. In this paper, we propose a different method to achieve feature extraction and tracking. For feature extraction, our method simply requires the user to label a feature on two slices in the time-varying data, and then, it generates a set of optimized Gaussian mixture model rules that can be used to automatically extract the feature at each time step in the time-varying data. Based on the extracted feature at each time step, our tracking method can create a global tracking graph that will record all possible tracking information of this feature across all time steps and thus achieve a global feature tracking. To demonstrate the effectiveness of our method, we applied several time-varying datasets from scientific simulations to it. Furthermore, to validate our method, we both qualitatively and quantitatively compared its feature tracking results against the ones from two state-of-the-art feature tracking techniques by referring to the ground truth. The experiment results showed that our method could generate the most accurate feature tracking results than the two compared state-of-the-art techniques.