Compact representation of redundant signals using the wavelet packet transform

We represent a novel approach to efficient representation of redundant signals that is based on statistical interdependence of wavelet packet (WP) coefficients of such signals. The WP coefficients are divided into a set of coefficients exhibiting high information content and a set containing the remaining coefficients with low information content. The WP coefficients with high information content can help to predict other coefficients reasonably well whereas the coefficients of the latter set contain much redundancy and can be predicted fairly well using the first set. This method leads to the reduced energy WP coefficients with a low information content which corresponds to a narrowed probability density function of these coefficients and yields a lower bit rate when using entropy coding or vector quantization. The algorithm has been successfully applied to several single lead electrocardiogram (ECG) signals from the widely used MIT-BIH arryhtmia database.