Mario Lasseck
- erosion, and dilation on the binary image (in that order) using a 4 by 4 neighbourhood.
- median filtering (this is not working at the moment, and messes up the removal of small objects somehow)
- remove small objects which have a size smaller than group size and a connectivity of 1
In this example of the process we can see the preprocessing done in the baseline which we are trying to reproduce (left), and the preprocessing done in our implementation (right).
| Our Method | Mario Lasseck |
|---|---|
 |
 |
 |
 |
 |
 |
EDIT: It makes more sense to do this in the time domain.
In order for this to work we need to know that computation of spectrograms is distributive over addition. That is, is the following true?
We know that DTF is distributive over addition. That is,
F{f1(x, y) + f2(x, y)} = F{f1(x, y)} + F{f2(x, y)}
Since computing a spectrogram is simply running a Short-Time Fourier Transform (STFT) over an input signal, it should be the case that
spectrogram(x1) + spectrogram(x2) = spectrogram(x1 + x2).
However, this needs to be more rigourous for a formal statement. We will confirm this empirically using simple sinusoids.