Modelling of visual feature derivation in the VizIR framework

If visual information retrieval should make further progress, it will be necessary to identify new ways to derive visual properties from higher levels of understanding than the pixel level (e.g. from low-level features). The paper outlines the implementation of modelling of feature hierarchies in the visual information retrieval framework VizIR (free under GPL). The approach allows for the derivation of high-level features from low-level features by aggregation and localisation as well as semantic enrichment with additional knowledge. The technical implementation is based on the MPEG-7 structures for aggregation and specialisation.