An important problem in signal analysis is to define a general purpose signal representation which is well adapted for developing pattern recognition algorithms. In this paper we will show that such a representation can be defined from the position of the zero-crossings and the local energy values of a dyadic wavelet decomposition. This representation is experimentally complete and admits a simple distance for pattern matching applications. It provides a multiscale decomposition of the signal and at each scale characterizes the locations of abrupt changes in the signal. We have developed a stereo matching algorithm to illustrate the application of this representation to pattern matching.