Using Raster Sketches for Digital Image Retrieval

James D. Carswell, Dublin Institute of Technology

Document Type Theses, Ph.D

Ph.D. Dissertation; Department of Spatial Information Science and Engineering; The University of Maine; May, 2000


This research addresses the problem of content-based image retrieval using queries on image-object shape, completely in the raster domain. It focuses on the particularities of image databases encountered in typical topographic applications and presents the development of an environment for visual information management that enables such queries. The query consists of a user-provided raster sketch of the shape of an imaged object. The objective of the search is to retrieve images that contain an object sufficiently similar to the one specified in the query. The new contribution of this work combines the design of a comprehensive digital image database on-line query access strategy through the development of a feature library, image library and metadata library and the necessary matching tools. The matching algorithm is inspired by least-squares matching (lsm), and represents an extension of lsm to function with a variety of raster representations. The image retrieval strategy makes use of a hierarchical organization of linked feature (image-object) shapes within the feature library. The query results are ranked according to statistical scores and the user can subsequently narrow or broaden his/her search according to the previously obtained results and the purpose of the search.