Work item
Web query interface and multiple feature fusion
Description
In those weeks, Web query interface is refined. The search results are based on two visual features that are GCM and Gabor.
An average fusion approach is used then to obtain final search scores for photos in data set.
For annotation part, the system currently provides three types of strategies, which are representative tags, TDIDF tags and frequent tags.
Observation
The suggested annotations or tags for a given photo contain a lot of noises. Due to the semantic gap existing in current CBIR systems, the suggested tags are somehow unrelated to the purpose of user.
A possible approach to solve this problem is going to rely on a given search term using as a preliminary term.
Next stage it is worth trying inducing Ontologies from tags. Ontology may help user to find and annotate photos more appropriate.
