Abstract

A feature detection algorithm for detecting singularity-like features in a one-dimensional signal has been developed. It is based on the zero-crossings wavelet representation introduced by Mallat [1], and involves a variance-weighted cross-correlation of the appropriate wavelet transform of the signal in the various frequency bands with an averaged template characterizing the particular feature. The algorithm was developed for the purpose of automatic feature detection in averaged evoked potentials. It successfully detects waves I to V in the auditory evoked potential and the N20 and P25 peaks in the median nerve stimulated somatosensory evoked potential, in normal and delayed-latency cases. The performance of the algorithm is plotted as a function of the signal-to-noise ratio. A priori knowledge of the expected feature amplitudes and latencies is not necessary, but may be incorporated into the model. By feeding the feature amplitude and latency estimates back into the algorithm as a priori knowledge, it is configured to track the changing characteristics of a feature over time.