Date of Award:
Doctor of Philosophy (PhD)
Connectedness is an important topological property and has been widely studied in digital topology. However, three main challenges exist in applying connectedness to solve real world problems: (1) the definitions of connectedness based on the classic and fuzzy logic cannot model the “hidden factors” that could influence our decision-making; (2) these definitions are too general to be applied to solve complex problem; and (4) many measurements of connectedness are heavily dependent on the shape (spatial distribution of vertices) of the graph and violate the intuitive idea of connectedness.
This research focused on solving these challenges by redesigning the connectedness theory, developing fast algorithms for connectedness computation, and applying the newly proposed theory and algorithms to solve challenges in real problems.
The newly proposed Neutro-Connectedness (NC) generalizes the conventional definitions of connectedness and can model uncertainty and describe the part and the whole relationship. By applying the dynamic programming strategy, a fast algorithm was proposed to calculate NC for general dataset. It is not just calculating NC map, and the output NC forest can discover a dataset’s topological structure regarding connectedness.
In the first application, interactive image segmentation, two approaches were proposed to solve the two most difficult challenges: user interaction-dependence and intense interaction. The first approach, named NC-Cut, models global topologic property among image regions and reduces the dependence of segmentation performance on the appearance models generated by user interactions. It is less sensitive to the initial region of interest (ROI) than four state-of-the-art ROI-based methods. The second approach, named EISeg, provides user with visual clues to guide the interacting process based on NC. It reduces user interaction greatly by guiding user to where interacting can produce the best segmentation results.
In the second application, NC was utilized to solve the challenge of weak boundary problem in breast ultrasound image segmentation. The approach can model the indeterminacy resulted from weak boundaries better than fuzzy connectedness, and achieved more accurate and robust result on our dataset with 131 breast tumor cases.
Xian, Min, "Neutro-Connectedness Theory, Algorithms and Applications" (2017). All Graduate Theses and Dissertations. 6527.
Copyright for this work is retained by the student. If you have any questions regarding the inclusion of this work in the Digital Commons, please email us at .