A Latent Image Semantic Indexing Scheme for Image Retrieval on the Web

Abstract

AbstractIn this paper, we present a novel latent image semantic indexing scheme for efficient retrieval of WWW images. We present a hierarchical image semantic structure called HIST, which captures image semantics in an ontology tree and visual features in a set of specific semantic domains. The query algorithm works in two phases. First, the ontology is used for quickly locating the relevant semantic domains. Second, within each semantic domain, the visual features are extracted, and similarity techniques are exploited to break the “dimensionality curse”. The target images can then be efficiently retrieved with high precision. The experimental results show that HIST achieves good query performance. Therefore, our method is promising in diverse Web image retrieval.KeywordsVisual FeatureImage RetrievalQuery ImageRelevance FeedbackSalient ObjectThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.