Department of Computer Science
The correlation between semantic visual similarity and ontology-based concept similarity in effective web image search
This paper compares the correlations between visual similarity of real-world images and different ontology-based concept similarity in order to find a novel measurement of the relationship between semantic concepts (objects, scenes) in visual domain besides low level feature extraction. For selected concept pairs, we compute their visual similarity and co-occurrence, which is represented by our Probability-based Visual Distance Model (PVDM). Rather than high computational cost of object recognition, by employing the ontology-based concept similarity into query expansion and filtering, the semantic image search and retrieval precision will be much higher. Furthermore, the latent topic will be mapped into images so that users are possible to retrieval the images with satisfying visual characteristic of the target concept. © 2012 Springer-Verlag Berlin Heidelberg.
image search, ontology, query expansion, visual similarity
Source Publication Title
APWeb 2012 International Workshops: SenDe, IDP, IEKB, MBC, Kunming, China, April 11-13, 2012. Proceedings
Link to Publisher's Edition
Leung, Clement. H.C., and Yuanxi Li. "The correlation between semantic visual similarity and ontology-based concept similarity in effective web image search." APWeb 2012 International Workshops: SenDe, IDP, IEKB, MBC, Kunming, China, April 11-13, 2012. Proceedings (2012): 125-130.