|
Ph.D. Defense Wednesday, May 30, 2007 ENG II – Room 208 10:00AM Title: Learning in Image and Geo-spatial Databases Abstract: This dissertation presents learning approaches to handle image and geo-spatial database problems, on the topic of both classification and indexing. For classification, a hybrid of transductive learning and feature synthesis learning method is designed. It overcomes both the labeling problem and the curse of dimensionality problem. Experimental results show it has better performance than the well-known SVM and other existing transductive learning methods and it is computational efficient in the testing phase. This dissertation also includes the design of an indexing structure, which is primarily for handling uncertain spatial databases. This Gaussian mixture based indexing technique can handle both the uncertain data and the uncertain queries. It outperforms variants of R-tree and it has a great real-world application. Moreover, it is also applicable to image databases. |
| < Prev | Next > |
|---|