Date of Award
Level of Access
Doctor of Philosophy (PhD)
Spatial Information Science and Engineering
Second Committee Member
Third Committee Member
Mary Kate Beard-Tisdale
Change detection is a topic of great importance for modern geospatial information systems. Digital aerial imagery provides an excellent medium to capture geospatial information. Rapidly evolving environments, and the availability of increasing amounts of diverse, multiresolutional imagery bring forward the need for frequent updates of these datasets. Analysis and query of spatial data using potentially outdated data may yield results that are sometimes invalid. Due to measurement errors (systematic, random) and incomplete knowledge of information (uncertainty) it is ambiguous if a change in a spatial dataset has really occurred. Therefore we need to develop reliable, fast, and automated procedures that will effectively report, based on information from a new image, if a change has actually occurred or this change is simply the result of uncertainty. This thesis introduces a novel methodology for change detection in spatial objects using aerial digital imagery. The uncertainty of the extraction is used as a quality estimate in order to determine whether change has occurred. For this goal, we develop a fuzzy-logic system to estimate uncertainty values fiom the results of automated object extraction using active contour models (a.k.a. snakes). The differential snakes change detection algorithm is an extension of traditional snakes that incorporates previous information (i.e., shape of object and uncertainty of extraction) as energy functionals. This process is followed by a procedure in which we examine the improvement of the uncertainty at the absence of change (versioning). Also, we introduce a post-extraction method for improving the object extraction accuracy. In addition to linear objects, in this thesis we extend differential snakes to track deformations of areal objects (e.g., lake flooding, oil spills). From the polygonal description of a spatial object we can track its trajectory and areal changes. Differential snakes can also be used as the basis for similarity indices for areal objects. These indices are based on areal moments that are invariant under general affine transformation. Experimental results of the differential snakes change detection algorithm demonstrate their performance. More specifically, we show that the differential snakes minimize the false positives in change detection and track reliably object deformations.
Mountrakis, Georgios, "Context-Specific Preference Learning of One Dimensional Quantitative Geospatial Attributes Using a Neuro-Fuzzy Approach" (2004). Electronic Theses and Dissertations. 569.
Files over 10MB may be slow to open. For best results, right-click and select "save as..."