> I'm trying to classify situations by their wavelet spectra, and I'm
> looking for a good way to characterize each spectrum. What ways are
> there to extract a small number of characteristic values from a large
> multiscale analysis result that allow recognizing similar situations?
Wavelet decompositions (and other transforms in general) are typically
used to attain good sparsity in the transform spectrum, i.e. having
*most* of the signal energy in as few as possible high magnitude
coefficients. Depending on your situation, you can take advantage of
this fact and represent classes of signals by their "representative"
high magnitude coefficients. Also, if you use a dyadic wavelet
transformation, you obtain multiresolution approximations and detail
coefficients. In typical situations, the approximation ones are all
preserved (and in case of full-dyadic decomposition, there is only one
approximation coefficient, the DC coefficient).
Kostadin
Reply by Andor●April 24, 20072007-04-24
Andreas Steffen wrote:
> Hello!
>
> I'm trying to classify situations by their wavelet spectra, and I'm
> looking for a good way to characterize each spectrum. What ways are
> there to extract a small number of characteristic values from a large
> multiscale analysis result that allow recognizing similar situations?
That depends. What characterizes your situations?
Regards,
Andor
Reply by Andreas Steffen●April 24, 20072007-04-24
Hello!
I'm trying to classify situations by their wavelet spectra, and I'm
looking for a good way to characterize each spectrum. What ways are
there to extract a small number of characteristic values from a large
multiscale analysis result that allow recognizing similar situations?
Thanks
Andreas Steffen