In this paper, a novel multi-scale, statistical approach for natural image representation is presented. The approach selects, at different scales, sets of features that, represent exclusively the most typical visual elements of several natural scene categories, disregarding other non-characteristic, clutter, elements. Such features provide also a robust image visual signature, useful for scene understanding, image classification and retrieval. The approach lies upon a structured generative model efficiently trained through variational learning. Results regarding image classification and retrieval prove the goodness of the approach.
Unsupervised Learning of saliency concepts for image classification and retrieval
PERINA, Alessandro;CRISTANI, Marco;MURINO, Vittorio
2008-01-01
Abstract
In this paper, a novel multi-scale, statistical approach for natural image representation is presented. The approach selects, at different scales, sets of features that, represent exclusively the most typical visual elements of several natural scene categories, disregarding other non-characteristic, clutter, elements. Such features provide also a robust image visual signature, useful for scene understanding, image classification and retrieval. The approach lies upon a structured generative model efficiently trained through variational learning. Results regarding image classification and retrieval prove the goodness of the approach.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.