Segmentation d'un signal experimental a partir de descripteurs morphologiques application a l'electroencephalogramme

Abstract Many signals can be defined as a background interrupted by events. Our studied signal, the electroencephalogram, belongs to a special field of experiments in which an expert interprets these events. To carry out automatically this interpretation we have to deal with different approaches, e.g. signal processing, pattern recognition and artificial intelligence. The aim is to research a special pattern with hierarchical cancelling of all the parts of the signal which cannot contain the pattern. To discriminate we have to define what we call a disparity and an analogy. We specify them in terms of morphological structures represented with syntactical rules. Here we present a method of segmentation based on local morphological differences which appear on the signal. We code these differences. The organisation of disparities allow us to separate the parts of the signal which cannot be considered as the background signal. Among these separated parts we research a special type of pattern. For this purpose we use syntactical rules that we built up on a learning set of pattern. These rules concern the elements of the code used for the disparities. We applied this study to electroencephalographic signals and to typical epileptic pattern as complex spikes and waves.