Inverse transform



next up previous
Next: Morlet wavelet and Up: No Title Previous: Transform of modulated

Inverse transform

As in the case of Fourier transforms, the wavelet maps suggest that the signal is made of various contributions, and in fact that it may be reconstructed from the wavelet coefficients. Unfortunately, the value of the signal at one time is not a simple superposition of simultaneous values from one vertical line of the wavelet map: rather, contributions from the neighborhood of this time must be used. In fact, we have a convolution (at each duration) of the wavelet coefficients with the analyzing wavelet, followed by an integration over all durations. On the Fig. 15, we see that the signal is reconstructed faithfully, except in this particular case for the highest frequency part of the noise (a shortcoming that could be remedied by extending the wavelet map toward shorter durations).




Figure 15: Reconstruction of the original signal (top) from its wavelet map (middle) is shown at the bottom. Some loss of short-duration fluctuations is seen.

Analytically, the inverse transform for the Mexhat is given by



This integral implies that the continuous wavelet transform is available over all durations and all times. In experimental practice, discrete values of the signal and of the wavelet map will be available, and the ensuing errors are studied within the theory of frames: see Daubechies's book for an introduction.



next up previous


Jacques Lewalle
Mon Nov 13 10:51:25 EST 1995