The University of Michigan School of Information Multi-mode Image Retrieval Group is engaged in research to develop and evaluate image retrieval techniques that utilize both visual and textual cues to query image databases. We are developing these content-based image retrieval methods to enable users to search for images based on image features, such as color, shape, and texture. In addition, our work will combine these methods with traditional text-based retrieval methods, including keyword searching and browsing. One of the primary goals of this work is to develop retrieval techniques that focus on the needs of generalist or naive users.

We are interested in the strategies that users employ in a search and how they think through an image search. Do they have a mental image of what they’re searching for when they initiate the search? What does that mental image look like and how does it compare with the image that a user eventually selects? How does the image search differ in looking for an abstract concept as opposed to a specific object? When do users use browsing versus a direct search? What are the criteria used to select an image from a retrieved set?

Our research team has developed an image retrieval system which allows users to search for images by browsing through a list of subject terms, by keyword search, and by looking for an image based on its image features - color, shape, or texture. Our image database consists of 1400 images in the area of earth and space science, and each image has been assigned keywords and descriptive information. We developed a broad organizational scheme for the browsing categories. We also developed search engines for both the textual and image feature searches.