Caixia Wang

Date of Award


Level of Access Assigned by Author

Open-Access Thesis

Degree Name

Master of Science (MS)


Spatial Information Science and Engineering


Anthony Stefanidis

Second Committee Member

Peggy Agouris

Third Committee Member

M. Kate Beard-Tisdale


With recent advances in sensor technology and digital image processing techniques, automatic image mosaicking has received increased attention in a variety of geospatial applications, ranging from panorama generation and video surveillance to image based rendering. The geometric transformation used to link images in a mosaic is the subject of image orientation, a fundamental photogrammetric task that represents a major research area in digital image analysis. It involves the determination of the parameters that express the location and pose of a camera at the time it captured an image. In aerial applications the typical parameters comprise two translations (along the x and y coordinates) and one rotation (rotation about the z axis). Orientation typically proceeds by extracting from an image control points, i.e. points with known coordinates. Salient points such as road intersections, and building corners are commonly used to perform this task. However, such points may contain minimal information other than their radiometric uniqueness, and, more importantly, in some areas they may be impossible to obtain (e.g. in rural and arid areas). To overcome this problem we introduce an alternative approach that uses linear features such as roads and rivers for image mosaicking. Such features are identified and matched to their counterparts in overlapping imagery. Our matching approach uses critical points (e.g. breakpoints) of linear features and the information conveyed by them (e.g. local curvature values and distance metrics) to match two such features and orient the images in which they are depicted. In this manner we orient overlapping images by comparing breakpoint representations of complete or partial linear features depicted in them. By considering broader feature metrics (instead of single points) in our matching scheme we aim to eliminate the effect of erroneous point matches in image mosaicking. Our approach does not require prior approximate parameters, which are typically an essential requirement for successful convergence of point matching schemes. Furthermore, we show that large rotation variations about the z-axis may be recovered. With the acquired orientation parameters, image sequences are mosaicked. Experiments with synthetic aerial image sequences are included in this thesis to demonstrate the performance of our approach.