Re: Definition of a (Gabor?) filter

You must be correct, because a similar approach for removing
illumination effects (or shadow) calculates differences of Gaussian
filtered image and originial image.

Yes, what I need to do is trying a face recognition system under
different lightning conditions.


On 23 Temmuz, 16:58, Matthias <ms.tre...@xxxxxxxxx> wrote:
On 22 Jul., 04:24, ist <say...@xxxxxxxxx> wrote:

The mentioned filter on the article is used to reduce effects of
illumination change. But I see that dog is rather used for edge
detection.

Yep. But it's actually an inverse DoG, as I said with a reponse
profile similar to a Gabor.
With "reducing the effects of illumination change" you mean that you
e.g. want to match two pictures under different illuminations? You can
use convolution with Gabors to measure Fourier energy at different
scales. I guess spectra would remain fairly constant under moderate
illumination changes, so you could use the spectrum for matching
pictures, if this is what the article is about.
Sounds like you work in vision science?


.