Efficient distribution-based feature search in multi-field datasets
Tzu‐Hsuan Wei, Chunming Chen, Jonathan Woodring, Huijie Zhang, Han‐Wei Shen · 2017
Local distribution search is used in query-driven visualization for identifying salient features. Due to the high computational and storage costs, local distribution search in multi-field datasets is challenging. In this paper, we introduce two high performance, memory efficient algorithms for searching for local distributions that are characterized by marginal and joint features in multi-field datasets. They leverage bitmap indexing and local voting to efficiently extract regions that match a target distribution, by first approximating search results and refining to generate the final result. The first algorithm, merged-bin-comparison (MBC), reduces the computation of histogram dissimilarity measures by clustering bins. The second algorithm, sampled-active voxels (SAV), adopts stratified sampling to reduce the workload for searching local distributions with large spatial neighborhoods. The efficiency and efficacy of our algorithms are demonstrated in multiple experiments.