This thesis focused on the analysis of electrocardiograms in view of developing new effective methods of classification of arrhythmias (a diagnostic tool) and automatic localisation of abnormal beats (monitoring tool) in real time in an ECG signal. The ECG signals are preprocessed and the extracted beats are compressed and then analysed using wavelet transform. The proposed classification method exploits specificities of the patient by doing a contextuel clustering of beats and using a database of annotated heart beats. The method uses also a similarity function to compare two given beats. The localisation method also uses the wavelet decomposition but operates only on a portion of available data (set of parsimony) to automatically detect in real time abnormal heart beats with the aid of a mask function. Both methods were tested on ECG signals from MIT-BIH arrhythmia database and good results have been obtained.
