The properties of two algorithms (zero-crossings algorithm and extrema algorithm) commonly applied to the automatic treatment of EEG signals are studied in full detail in order to make a comparison of their respective performances: quality of the signal restitution; reduction ratio; computation time; ability to discriminate electrophysiological states. . This study demonstrates that these algorithms although seeming rather similar at the outset.
actually have quite different fields of applicability. The algorithm of zerocrossings is well adapted to the analysis and discrimination of background activities, especially over long periods of time, at the cost of a loss of information relative to short duration phenomena (transients, actual waveform of the signal). The algorithm of extrema, in contrast, constitutes a pretreatment procedure, providing a reduced image of the primary signal without noticeable loss of information. It is thus better adapted to the analysis of short sequences or transient phenomena. It is also demonstrated that the properties of these algorithms make it possible to generalize their applications to electrophysiological signals other than EEG, EOG for example. Lastly, they permit application to continuous phenomena of all the methods of analysis of point processes usually devoted to discrete phenomena.