Re: sequence segmentation

[Peter L. Soendergaard]

Hi Pascal,

On 07/09/2013 08:36 PM, CdeMills wrote:
> Hello,
>
> I have an acquisition sequence which can be described as an initial
> transient, followed by segments with different noise level. Moreover,
> isolated spikes are also present. I tried automatic segmenting. It can be
> seen as a classification according to the density of points whose distance
> to the median value is higher to some threshold. But it is not easy to
> decide wheter or not a point is a simple isolated outlier or belongs to a
> group of points where the local noise is greater than expected. Any idea or
> pointer to algorithm ? Any octave toolbox to perform such segmenting ?
I don't have a solution, but perhaps a way of making it easier: Have you 
tried preprocessing your data through a transform that enhances the 
spikes and reduces the noise? A simple wavelet transform might improve 
the "contrast".

LTFAT can do it, it is not yet available as an Octave package (we are 
currectly changing some interfaces), but you can download it from 
Sourceforge. Just use something like "fwt(insig,'db4',8)" and try 
segmenting on the output.

Cheers,
Peter.