Image indexing based on visual content is an especially active and challenging field in image processing. Without any restriction on processed images, we indeed face contents which may be heterogeneous, ambiguous and also acquired in poor conditions. As difficult as it may appear, most of the time, this activity poses very few problems to human beings who always reach a quick classification decision, whichever the complexity of the original image.

An automated indexing system should, ideally, allow searching for concepts within a heterogeneous image and being able to detect their presence as well as their absence in a non-mutually exclusive way. Our first objective was to draw means of processing information from human perception which would put us into good conditions to make a successful classification. We also devised efficient shape features to help us in this classification task. Finally, we developed an efficient classification process that could adapt to these difficult conditions.