Recuperação de imagens na Web baseada em múltiplas evidências textuais

Lamarque SouzaTatiana A. S. CoelhoBerthier Ribeiro-Neto

The growing availability of image databases on the Web and the fact that available image search engines don´t satisfy the user information necessity suggest the study of new information retrieval algorithms. Current image search engines use image digital processing techniques, or use a retrieval process based on simple labels, such as the image ?le name. As a result, the task of ?nding the best images that satisfy the user information need continues to be a challenging problem. In a different form as occur in the current algorithms applied to the image search engines, we understand that most of the sources of evidence that are available in the HTML document that contains an image should be used to retrieve and rank images in the answer set of the user query. In this paper, we explore various alternatives for indexing images, and we propose a Bayesian framework which allows combine them to obtain an improved quality ranking. We show that this framework provides effective mechanism for searching and ranking images on the Web. We also show that many sources of textual evidences improve meaningfully the answers quality.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: