Isabella Gagliardi , CNR, Italy
Bruna Zonta , CNR, Italy
Short Paper: Short Paper 6: Looking for Content
The retrieval of visual information from large archives is a need shared by many application domains. In some environments, such as that of a community's cultural heritage, the pertinence and quality of the information are issues of primary importance. Retrieval based on low-level features does not guarantee these properties: the image alone of a cultural artifact cannot convey all the necessary information; it must be completed by a description, and by its historical and environmental collocation, that is, by an accompanying text. Unfortunately, systems that retrieve images through texts in a completely unsupervised manner have been seen to produce results that are not only unsatisfactory, but difficult for even experts in the sector to understand. We present here Quicklink, a system that retrieves images similar to a query image in large Cultural Heritage web archives by dynamically matching their textual descriptions (usually catalog cards), adapts its behavior to user requests, and presents the retrieval results in HTML pages, where the images are ordered according to their degree of similarity. The Quicklink system, created as a stand-alone application has now been extended to an Internet application that can adapt the system's standard behavior to the user's needs. Briefly, when, the system presents the retrieved images in response to a query, the user can apply the mechanism of relevance feed-back to progressively refine the results by indicating their pertinence or non-pertinence. Following the user's choices, the system temporarily increases or diminishes the weights of the terms involved, and then reapplies the similarity algorithm. The catalogues of the Faenza International Museum of Ceramics form the principal part of the test bed. In the paper we shall present the web application, describing in detail the similarity algorithm, the adaptation algorithm, and the overall evaluation of the system's feasibility and efficiency on the Internet.