The SubpixelLocalization class is based on quadratic fitting of the peak
it tries to localize. To retrieve an estimate of the position, we
compute the gradient and the hessian at the peak integer location.
To compute the gradient and hessian, we use a randomAccess on the peak
image, and iterate through dimensions.
It turns out that the randomAccess position was not restored at the peak
center after dealing with a dimension. Therefore, the next dimensions
were calculated off-peak.
tpietzsch
commented
9 years ago
The SubpixelLocalization class is based on quadratic fitting of the peak it tries to localize. To retrieve an estimate of the position, we compute the gradient and the hessian at the peak integer location.
To compute the gradient and hessian, we use a randomAccess on the peak image, and iterate through dimensions.
It turns out that the randomAccess position was not restored at the peak center after dealing with a dimension. Therefore, the next dimensions were calculated off-peak.
Fixes #12.
Signed-off-by: Jean-Yves Tinevez jean-yves.tinevez@pasteur.fr