Image retrieval by composition of regions
Résumé
We present an image retrieval system called "STRICT". Its design is based on a formulation of fuzzy similarity measures that is derived from psychological considerations and that leads to an intuitive formulation of the similarity. We use these measure to evaluate a region to region visual similarity score. Then, we aggregate these scores in different ways depending on the purpose of the request. By aggregating basic scores such as region to region similarity values we form various kinds of composite regional queries. These queries can cover several different user needs such as scene retrieval, multiple aspects retrieval and object filtering.