Synthetic aperture sonar (SAS) has become a well developed imaging technique for imaging shallow water environments. Aperture synthesis provides high along-track resolution imagery, with range independent resolution. However, mapping of the seafloor using traditional SAS is limited to a two-dimensional surface. To provide the third dimension (height), an interferometric synthetic aperture sonar (InSAS) is formed, comprising of two or more vertically displaced hydrophone arrays. Each of the interferometric receiver datasets are processed using standard SAS algorithms, with motion compensation and corrective processing applied equally to each channel, preserving the underlying interferometric time delays. By then estimating the time delay of the incoming wavefronts across the interferometric receiver array, the height of the seafloor can be inferred from the side-scan geometry of the system. The InSAS approach is similar to the radar equivalent (InSAR), however, significant differences in geometry and medium properties limit the applicability of InSAR algorithms to the sonar equivalent. A height estimate from interferometric data is formed by estimating the time difference between the receiver elements of the interferometric array. Therefore, for an accurate estimate of the time-delay, the signals of the receivers must contain significant 'common' information. Presented in this thesis is an analysis of coherence as applicable to an InSAS system. The coherence of an InSAS system can be decomposed into five 'coherence components': additive acoustic noise, footprint misalignment, baseline decorrelation, temporal decorrelation, and processing noise. Of these, it is shown footprint misalignment has the greatest effect for an InSAS system if it is not corrected for. The importance of maintaining high coherence between the receiver channels is presented; small losses in coherence from the ideal of unity will have a significant impact of the accuracy of the resulting height estimate. To reduce the sensitivity of the height accuracy losses, multiple estimates of the height can be formed from independent 'looks' of the scene. Combining all these estimates into one height estimate is shown to significantly improve the height estimate. The design and signal processing of an InSAS system is of high importance to the generation of high accuracy height estimates of the seafloor. Several parameters of design are explored, in particular the effect of aperture sampling. Low along-track aperture sampling rates are shown to cause a significant decrease in signal coherence, caused by the generating of 'grating lobes' from the synthetic aperture processing. Substantial improvements can be made by careful selection of transmitter and receiver element sizes, relaxing the requirements of a highly sampled aperture. An analysis of interpolation schemes on interferometric quality is also presented. The effect of footprint misalignment can be reduced by first resampling the data from each receiver onto a common ground-plane. However, this requires prior knowledge of the seafloor height, an unknown parameter before an interferometric height estimate is made. One possible method to form an initial height estimate is through the use of belief propagation, a technique applied from the field of stereo imaging. Belief propagation is used to estimate an initial height surface, albeit at discrete height intervals. This initial low resolution height surface can then be used to remap the data, partially eliminating the detrimental effects of footprint misalignment. The combination of all the independent estimates of the scene can be combined using maximum likelihood estimation. This framework allows the individual estimates to be combined into one overall cost function. Searching of the cost function for minimum cost yields a single interferometric time-delay estimate, from which a single height estimate can be inferred. This framework allows looks formed from many different sources to be combined, including multiple imaging frequency bands, and the use of more than one interferometric pair of receivers.

This thesis examines property generalization among concepts. Its primary objective is to investigate the hypothesis that the more central a feature for a concept, the higher its generalizability to other concepts that share a similar structure (features and dependencies). Its secondary objectives are to examine the relative contributions of feature centrality and feature variability in property induction, whether centrality offers a domain-general or a domain-specific constraint, and whether centrality can operate under conditions of vagueness. Experiments 1 and 2 addressed the centrality hypothesis with centrality measured, whereas Experiments 3 to 14 and 17 with centrality manipulated. Relative feature centrality was manipulated as follows: from a single-dependency chain (Experiments 3 to 7), from the number of properties that depended upon a feature (Experiments 8 to 11 and 17), and from the centrality of the properties that depended upon the critical features (Experiments 12 to 14). The results support the centrality hypothesis. Experiments 12 to 16 addressed the relative contributions of centrality and variability in property induction. Experiments 12 to 14 pitted a central and variable property against a less central and less variable property in judgments of frequency and inductive strength. The results suggest that property induction depends on centrality rather than frequency information, and that centrality can bias the perception of frequency (although the latter results were not clear-cut). Experiments 15 and 16 pitted centrality against variability in information seeking. The results show that centrality information is sought more often than variability information to make an inference, especially amongst dissimilar concepts. Experiments 1 to 16 used animal categories. Experiment 17 examined the centrality hypothesis with artifact categories. The results show centrality effects. Taken together, the Experiments suggest that centrality offers a domain-general constraint. Experiments 5, 8 to 11, and 17 left the properties that depended upon a candidate feature unspecified. A centrality effect was still obtained. The results suggest that centrality can operate under conditions of vagueness. The results are discussed in terms of theories of conceptual structure and models of category-based inference. A model to capture the present findings is also sketched.

Thesis (Ph.D.) - University of Glasgow, 2003. / Ph.D. thesis submitted to the Department of Cardiovascular and Medical Sciences, Faculty of Medicine, University of Glasgow, 2003. Includes bibliographical references. Print version also available.

Thesis (Ph. D.)--University of Iowa, 2008. / Thesis supervisor: Milan Sonka. Includes bibliographical references (leaves 145-150).

Thesis (Ph. D.)--University of Texas at Austin, 2004. / Supervisors: Thomas E. Milner and H. Grady Rylander, III. Vita. Includes bibliographical references.

Thesis (Ph. D.)--University of Iowa, 2009. / Thesis supervisors: Sonka, Michael D. Abràmoff. Includes bibliographical references (leaves 76-80).

Thesis (Ph. D.)--University of Washington, 2000. / Vita. Includes bibliographical references (leaves 200-205).

Thesis (Ph. D.)--University of Oregon, 1999. / Typescript. Includes vita and abstract. Includes bibliographical references (leaves 131-135). Also available for download via the World Wide Web; free to University of Oregon users. Address: http://wwwlib.umi.com/cr/uoregon/fullcit?p9955916.

Thesis (M.S.)--Marquette University, 2009. / Access available to Marquette University only. Joseph Carroll, Kristina Ropella, Taly Gilat-Schmidt, Advisors.

No description available.