| This thesis presents new methods for integrating spatial and feature information in order to improve systems for image retrieval, analysis and compression. In particular, this thesis develops and demonstrates an integrated spatial and feature query paradigm for image retrieval, a general framework for extracting spatially localized features from images using histogram and binary set back-projections, a system for image compression and analysis based upon a joint-adaptive space and frequency (JASF) graph, and a representation of texture based upon spatial-frequency energy histograms.; These techniques are similar in their exploitation of images as two-dimensional, non-stationary signals. The visually important information within images is often confined to spatially localized regions, or is represented by the spatial arrangements of these regions. By developing processes that analyze and represent images in this way, we improve our capabilities to develop powerful content-based image retrieval and image compression systems.; This thesis examines several image system applications, such as image and video storage and retrieval, content-based image query, adaptive image compression and image analysis. We demonstrate that by using the proposed integrated spatial and feature approaches, we improve image system performance in each application over non-integrated methods. |
