Comparison and Tracking Methods for Interactive Visualization of Topological Structures in Scalar Fields

• Main Author: Saikia, Himangshu
• Format: Doctoral Thesis
• Language: English
• Published: KTH, Beräkningsvetenskap och beräkningsteknik (CST) 2017
• Subjects: topology; scalar fields; merge tree; tree comparison; tracking; similarity search; Computer Science; Datavetenskap (datalogi)
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-216375; http://nbn-resolving.de/urn:isbn:978-91-7729-580-8

Scalar fields occur quite commonly in several application areas in both static and time-dependent forms. Hence a proper visualization of scalar fieldsneeds to be equipped with tools to extract and focus on important features of the data. Similarity detection and pattern search techniques in scalar fields present a useful way of visualizing important features in the data. This is done by isolating these features and visualizing them independently or show all similar patterns that arise from a given search pattern. Topological features are ideal for this purpose of isolating meaningful patterns in the data set and creating intuitive feature descriptors. The Merge Tree is one such topological feature which has characteristics ideally suited for this purpose. Subtrees of merge trees segment the data into hierarchical regions which are topologically defined. This kind of feature-based segmentation is more intelligent than pure data based segmentations involving windows or bounding volumes. In this thesis, we explore several different techniques using subtrees of merge trees as features in scalar field data. Firstly, we begin with a discussion on static scalar fields and devise techniques to compare features - topologically segmented regions given by the subtrees of the merge tree - against each other. Second, we delve into time-dependent scalar fields and extend the idea of feature comparison to spatio-temporal features. In this process, we also come up with a novel approach to track features in time-dependent data considering the entire global network of likely feature associations between consecutive time steps.The highlight of this thesis is the interactivity that is enabled using these feature-based techniques by the real-time computation speed of our algorithms. Our techniques are implemented in an open-source visualization framework Inviwo and are published in several peer-reviewed conferences and journals. === <p>QC 20171020</p>