| Summary: | The advent of the World Wide Web and digital photography has led to a phenomenal increase in the number and complexity of stored images. Accordingly, the ability to browse and retrieve images based upon image content is of rapidly growing importance. The goals of this research project are to develop a Content-Based Image Retrieval (CBIR) system that combines dynamic, user-driven search capabilities with artificial intelligence techniques and to examine the system's effectiveness. The experimental method will be used to test the specific hypotheses and various research questions proposed in this research project. All of the experiments will be conducted using a CBIR prototype system that incorporates intelligent User Interface Agents (UIA). The UlAs will use both neural networks and an expert reasoning system. The actual experiments will be conducted using a task-oriented approach, with both descriptive and analytical statistics used to assess the results. In addition, a new evaluation CBIR metric will be proposed and applied. It is expected that this research will benefit CBIR research and CBIR system development by: 1) demonstrating the effectiveness of providing users with an interface that allows them to sketch an image, provides a relevance feedback mechanism that is based on providing similar images, and offers query refinement suggestions; 2) presenting a reusable modular design approach that can be used to create CBIR systems; 3) showing how AI techniques, particularly intelligent User Interface Agents, can be used effectively in CBIR systems; 4) proposing a "standard" CBIR user interface; and 5) proposing a new CBIR evaluation metric. The results of this research project should advance the current state of CBIR in that it designs, implements and evaluates an interactive CBIR system that uses image input and incorporates both the user's interactive guidance and artificial intelligence techniques to access images.
|
|---|
