A multi-scale fuzzy classification by knn. Application to the interpretation of aerial images

Proposes an algorithm of multi-scale fuzzy classification by knn, used on complex data (multimodal, various size classes...). The method uses potential functions, corresponding to membership degrees of classes. It introduces a scale factor, specific to each class, to weight these functions. This multi-scale analysis improves the learning and the system performances, by making the potentials consistent over the feature space. The analysis adapts itself, especially to small size classes usually neglected in classification systems