Experiments with Clustering the Collection at ImageCLEF 2007

We present our participation in the 2007 ImageCLEF Ad-hoc photographic retrieval task. Our first participation in this year’s imageCLEF comprised six runs. The main purpose of three of these runs was to evalute the text and visual retrieval tools as well as their combination in the context of the given task. The other purpose of our participation is to experiment with applying clustering techniques to this task, which has not been done frequently in previous editions of ImageCLEF AD-hoc task. We use the preclustered collection to augment the search results of the retrieval engines. For retrieval we used two publicly available libraries; Apache Lucene for text and LIRE for visual retrieval. The clustered-augmented results reduced slightly the precision of the initial runs. While the aspired results have not yet been achieved, we note that the task is useful in assessing the validity of the clusters.