Output details

Concept-based indexing of annotated images using semantic DNA

The key innovation is the unsupervised concept-based image indexing technique, which uses a lexical ontology to extract a semantic signature called ‘semantic chromosome’ from image annotations. The approach has been implemented on an open access picture library FotoLibra (http://www.fotolibra.com/); it demonstrated better accuracy than the unsupervised algorithms submitted to the 2007 Semeval competition (http://www.senseval.org/). The research has been reported internationally through invited lectures and talks (2010-2012) in Japan (Tetsuo Sawaragi, sawaragi@me.kyoto-u.ac.jp), China (Juanzi Li, ljz@keg.cs.tsinghua.edu.cn), Hong Kong (WB Lee, WB.Lee@inet.polyu.edu.hk), France (Cecilia Zanni-Merk, cecilia.zanni-merk@insa-strasbourg.fr) and others. It has triggered exchanges of MSc and PhD students with Hong Kong and Strasbourg.