An Extended Vector Space Model for Content Based Image Retrieval

This paper describes participation of Dokuz Eylul University to the ImageCLEF2009Med task. This year, we proposed a new model for content-based image retrieval combining both textual and visual information in the same space. It simply extends traditional vector space model of text retrieval with visual terms. The proposed model also supports to close the semantic gap problem of content-based image retrieval. Experiments showed that our proposed system improves the performance of textual retrieval methods by adding visual terms. The proposed method was evaluated on the ImageCLEFmed 2009 dataset and it was ranked the best performance among the participants in automatic mixed retrieval including both text and visual features.