Comparison of Two Similarity Criterions in Shape Classification of Transient Signals: Application to High Resolution Electrocardiography

Abstract

In the analysis of recurrent time series the estimation of repetitive transient signals in currently achieved by the well known averaging technique. The efficiency of this technique depends on the accuracy of the alignement process and on the constancy in shape and width of the averaged signals. A typical example is the Highly Amplified ECG signal. In previous works we showed the advantage of selective averaging based on shape classification. The similarity criterion used in such a classification was derived from the Distribution Function Method. In this paper we compare this criterion to an other one using correlation, first by simulation and secondly on real ECG signals. The ability for the two approaches to give significantly different classes is studied.