Learning spatial relationships in hand-drawn patterns using fuzzy mathematical morphology

We introduce in this work a new approach for learning spatial relationships between elements of hand-drawn patterns with the help of fuzzy mathematical morphology operators. Relying on mathematical morphology allows to take into account the actual shapes of hand-drawn patterns when modeling their spatial relationships, and thus to cope with the variability of handwriting signal. Extension of mathematical morphology to the fuzzy set framework further allows to handle imprecision of handwriting and to deal with the ambiguity of spatial relationships. The novelty lies in the generative aspect of the models we propose, in the sense that they can exhibit the region of space where the learnt relation is satisfied with respect to a reference object, and can thus be used for driving structural analysis of complex patterns. Experiments over on-line handwritten data show their performance, and prove their ability to deal with variability of handwriting and reasoning under imprecision.