your mail

Bruno Postle bruno at postle.net
Wed Oct 22 10:57:16 BST 2003


On Wed 22-Oct-2003 at 07:17:58AM +0200, Kai-Uwe Behrmann wrote:
> Am 22.10.03, 00:11 +0100 schrieb Bruno Postle:
> 
> > Helmut had some reasoning for this that I can't remember (I can look
> > it up).
> 
> Would be interessting.

(14000 messages in that mbox)

I wrote:

> > As far as I can tell, and I could be wrong here, there is no way to
> > identify the correct number for any particular lens.
> >
> > There are two reasons for this:
> >
> > - The effect is scaled to the width of the image, so you need two
> >   different numbers for landscape and portrait photos.
> >
> > - It's additive rather than multiplicative (is that a word?), so dark
> >   images tend to get too much adjustment and light images get too
> >   little.

Helmut replied:

> This is how Radial Luminance works: A correction value depending
> on the radial distance R to the center is added/subtracted to/from
> each pixel value.  The dependence is given by a*R^2+b. The
> parameter set in the dialog is the difference in brightness
> bedween center and horizontal edge.  Eg using a value of 40,
> PTools adjusts the corecting polynomial so that center pixels are
> darkended by 20 and edge pixels are brightend by 20.
> 
> It should be mentioned that indeed this method does not exactly
> compensate vignetting. Depending on the exact cause for
> vignetting, this follows a power of cos(theta), and is
> "multiplicative" not "additive" for the intensities. However, the
> rgb-pixel values are gamma transformed intensity data, which turns
> multiplications (almost) into additions, and brightness
> adjustments are better done that way. Eg, the Photoshop Brightness
> adjustment is purely "additive".  Multiplications are perceived as
> contrast changes, which is not intended.
> 
> Specifying a gamma-corrected cos(theta) power law might be more
> exact, but not very practible. The polynomial is much easier to
> determine (by just entering the difference of center vs edge
> brightness) and corrects almost equally well.

-- 
Bruno


More information about the ptX mailing list