Image retrieval has been an active research area for many years, but two fundamental problems remain largely unsolved: 1) How best to learn users' subjective query concepts, and 2) How to measure perceptual similarity with significant accuracy. The first problem concerns the completeness of formulating a query concept, e.g., how to formulate a query such as ``animals,'' ``cathedrals,'' or ``aircraft.'' The second problem concerns search accuracy, i.e., given a learned query concept, how to find all images that match that concept.
To tackle these two fundamental problems and to ensure that our solutions are scalable, this project has four specific targets. First, we plan to develop novel active learning algorithms that quickly learn users' subjective query concepts (thoughts and intents) despite time and sample constraints. Second, we will design semi-automatic image annotation and annotation refinement methods for assigning semantic labels to images in order to support multimodality query-concept learning and information retrieval. Third, we will devise perceptual distance functions for improving accuracy of visual searches. For instance, once a query concept such as ``enemy vessels'' is learned, we want to find every matching object in the surveillance database, not missing any. Finally, we plan to conduct validation studies} on developed learning algorithms, using experimental data provided by colleagues at various institutions (including IBM research centers and Fine Arts Museums of San Francisco).
