The Use of QBIC Content-Based Image Retrieval System

Abstract

The fast increase in digital images has caught increasing attention on the development of image retrieval technologies. Content-based image retrieval (CBIR) has become an important approach in retrieving image data from a large collection. This article reports our results on the use and users study of a CBIR system. Thirty-eight students majored in art and design were invited to use the IBMŅs QBIC (Query by Image Content) system through the Internet. Data from their information needs, behaviors, and retrieval strategies were collected through an in-depth interview, observation, and self-described think-aloud process. Important conclusions are: (1) There are four types of information needs for image data: implicit, inspirational, ever-changing, and purposive. The types of needs may change during the retrieval process. (2) CBIR is suitable for the example-type query, text retrieval is suitable for the scenario-type query, and image browsing is suitable for the symbolic query. (3) Different from text retrieval, detailed description of the query condition may lead to retrieval failure more easily. (4) CBIR is suitable for the domain-specific image collection, not for the images on the Word-Wide Web. (Article content in Chinese with English abstract)

Keywords

• content-based image retrieval
• information seeking behavior
• users study