Author Topic: Trying to understand Vincent's "New Approach to the Combination of Broadband.."  (Read 2363 times)

Offline mattssporre

  • Newcomer
  • Posts: 20
    • Spur Photography
Dear all,

I am trying to understand Vincent's "New Approach to the Combination of Broadband and Narrowband Data" but is struggling  :( with the first step (the rest I think I understand - knock on wood).

http://www.pixinsight.com/tutorials/narrowband/theory/en.html

I know there are scripts that can combine NB with RGB data, but I still want to understand the way it is done.

Assume we work only with red data and that we have a Ha image and a Red broadband data. Let R(x,y) denote the value of pixel (x,y) in the Red BB image and let H(x,y) denote the value of pixel (x,y) in the Ha image.

In Vincent's method we first create a continuum map

C(x,y) = R(x,y)/H(x,y)

This should somehow represent the Red channel with the Ha part "flat fielded" away. However I believe some sort of normalization is needed. Consider the case

R(x,y) > H(x,y)    =>   C(x,y) = R(x,y)/H(x,y) >1

Values larger than one is outside of the PI range (has to be between 0 and 1), thus the continuum map will have "over exposed" areas, which mean that R(x,y) is not recovered by the operation R = C*H.

I see this in my manual attempts using pixelmath.

The reference to flat fielding (I think it was a post by Juan) made me recall that not even flat fielding is as simple as just dividing (image/flat). First the flat need to be normalized in two steps. The first step is to remove the threshold by subtracting a bias or a flatdark. The second step is to assume that a small region in the middle of the flat is free of vignetting (and other effects), computing the mean pixel value in that small region and then dividing the whole flat with that mean value.

I think this is what is missing and why I get the results explained above.

How is this normalization done in Vincent's method?

Note that just checking the "Rescale" option in pixelmath will not help. It re-scales the continuum map so that all pixel values stays between 0 and 1 alright, but the red image is still not recovered by C*H.

I might be completely stupid, missing the obvious. If so just point it out to me and I will henceforth go by the name Stupid   :)

BR
Matts