Home My Page Projects MPTK: The Matching Pursuit ToolKit
Summary Activity Forums Tracker Lists Docs News SCM Files

Forum: help

Monitor Forum | Start New Thread Start New Thread
RE: Individual Gabor atoms [ Reply ]
By: Jessica Beltran on 2014-02-25 19:15

scale question.png (69) downloads
I found a formula for Gabor atoms in this paper [1]

Given the next parameters for a single atom:

amp = 0.651673114873373;
chirp = 0;
freq = 0.117187500000000;
lenS = 128;
phase = 3.141472727003703;
pos = 78648;
windowoption= 0;
windowtype =9;

I convert them to seconds and Hz by using the samplig frequency
len in seconds=0.0464;
pos in seconds= 1.7834;
freq in HZ = 44100*.1171;

With this code in MATLAB I'm able to reconstruct a single atom

atom =zeros(100000,1); %length of the complete signal

for n=1:100000
atom(n,1) = (1/sqrt(len))*exp(-pi*(n/44100-pos)^2/len^2) * cos(2*pi*freq*(n/44100-pos)+phase);
atom = (1/norm(atom)) .* atom; %Normalization factor
figure; plot(atom);

Using the time shift and scale in number of samples will lead to the same result, just not divide n per 44100.

I'm using the frequency in Hz but I'm confused because in the paper it shown 2*pi*w.

Also, in the final vector, the atom does not consist of only 128 samples, but the other values are very close to zero.

[1]Environmental Sound Recognition With Time–Frequency Audio Features

RE: Individual Gabor atoms [ Reply ]
By: Jessica Beltran on 2014-02-24 22:05
I'm using MATLAB interface. I obtain these values for a single atom:

amp 0.651673114873373
chirp 0
freq 0.117187500000000
len 128
phase 3.141472727003703
pos 78648
windowoption 0
windowtype 9

As I'm using a sampling frequency (fs) of 44100 I can deduce that:
freq in Hz = freq*fs = 5.16KHz
time = pos/fs = 1.7834sconds

I still don't know how use the phase and the scale. As I know, the len parameter give me the number of samples or length of the atom, but this is not enough form me to reconstruct because I don't know how is the mother wavelet suppose to be.

RE: Individual Gabor atoms [ Reply ]
By: Christian Guilleminot on 2014-02-21 19:06
Instead of a binary book, request a book in XML. But I do not find in what units are the values​​.
For Gabor atoms envelope is a Gaussian curve.
The manual does not say which units are the measurements.
I asked these questions and I did not answer.
If anyone can publish this information would be useful.
De même pour des exemples de dictionnaires et leur usage

Individual Gabor atoms [ Reply ]
By: Jessica Beltran on 2014-02-20 21:53
Once I have decomposed a signal into atoms, I would like to have acces to the shape of each atom, this way I could calculate the energy or entropy of the atom.

Is there a way to extract each gabor wavelet individually?