High-retrieval precision in content-based image retrieval can be attained by adopting relevance feedback mechanisms. The user marks all the retrieved images as being either relevant or not, then the search engine exploits this relevance feedback to adapt the search to better meet user's needs. The main difficulties in exploiting relevance information are (a) the gap between user perception of similarity and the similarity computed in the feature space used for the representation of image content and (b) the availability of few training data (users typically label a few dozen of images. At present, SVM are extensively used to learn from relevance feedback due to their capability of effectively tackling the above difficulties. As the performances of SVM depend on the tuning of a number of parameters, in this chapter a different approach to relevance feedback is proposed. First, images are represented in the dissimilarity space made up of the dissimilarities from the set of relevant images. Then a relevance score is computed in terms of the distance from the nearest nonrelevant image, and the distance from the nearest relevant one. Images are ranked according to this score and the top k images are displayed. Reported results show that the performances of the proposed approach are comparable to the highest performances that can be attained by SVM by suitably tuning the learning parameters. © 2008 Springer-Verlag Berlin Heidelberg.

Instance-based relevance feedback in image retrieval using dissimilarity spaces

ROLI, FABIO
2008

Abstract

High-retrieval precision in content-based image retrieval can be attained by adopting relevance feedback mechanisms. The user marks all the retrieved images as being either relevant or not, then the search engine exploits this relevance feedback to adapt the search to better meet user's needs. The main difficulties in exploiting relevance information are (a) the gap between user perception of similarity and the similarity computed in the feature space used for the representation of image content and (b) the availability of few training data (users typically label a few dozen of images. At present, SVM are extensively used to learn from relevance feedback due to their capability of effectively tackling the above difficulties. As the performances of SVM depend on the tuning of a number of parameters, in this chapter a different approach to relevance feedback is proposed. First, images are represented in the dissimilarity space made up of the dissimilarities from the set of relevant images. Then a relevance score is computed in terms of the distance from the nearest nonrelevant image, and the distance from the nearest relevant one. Images are ranked according to this score and the top k images are displayed. Reported results show that the performances of the proposed approach are comparable to the highest performances that can be attained by SVM by suitably tuning the learning parameters. © 2008 Springer-Verlag Berlin Heidelberg.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11567/1084253
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 3
  • ???jsp.display-item.citation.isi??? 3
social impact