Re: [Gimp-user] double exposure

[yahvuu <yahvuu gmail com>]

Am 05.11.2012 20:01, schrieb Richard Gitschlag:
> PS:  Since linear RGB pixels values are typically gamma-encoded representations of some absolutely
> recorded brightness, I suppose technically for a double-exposure you would have to un-encode the
> gamma (e.g. apply a Levels adjustment of 0.45 or so), merge the two source layers using Addition
> blending, then re-encode (e.g. 2.2 Levels adjustment).  I did some experiments using two layers with
> black-to-white gradients rotated 90 degres from each other, and the result of those adjustments was
> indeed pretty close to the "Screen" blending mode.  That's pretty interesting.
this similarity is not by accident:

gamma curves can pretty closely resemble logarithms within the given value range and resolution:
http://www.wolframalpha.com/input/?i=x^0.8%2C+ln%28x%2B1%29%2Fln%282%29+from+x%3D0+to+1

Adding the logarithms of two layers and subsequent un-logarithmising
gives the product of the layers:
ln(a) + ln(b) = ln(a*b)
         <=>
a*b = exp( ln(a)+ln(b) )


The screen blend mode f(a,b) = 1-(1-a)*(1-b)  can be read as multiplication,
to be performed with inverted input layers and followed by inversion of the result.


best regards,
peter