Peter Heckert schrieb:
let me express it in more basic terms:
Colorspace conversion is a transformation of data.
linear transforms are commutative, nonlinear transforms are not commutative.
Blurring of photographic images happens in pysical reality, where light
intensities
are proportional to the number of photons/second.
Blurring and deblurring are linear convolutions, they are linear transforms.
Even lens-blur, that occurs in physical reality is a linear transform.
Therefore photographic images must be deblurred in a colorspace where
intensity values are approximately proportional to the number of photons.
This gives the best results.
Raw conversion into linear rgb is a linear transform.
Raw conversion into sRGB,Lab,AdobeRGB .. is a nonlinear transform.
Therefore this is ok:
Take photo -> convert raw into rgb (gamma=1)->deblur image->convert to
arbitrary output space
Take photo in sRGB->convert from sRGB to linear RGB(16bit)->deblur->
convert to output.
This is not ok: (gives bright halos)
Take photo->convert into sRGB->deblur
Take photo->convert into nonlinear colorspace (sRGB|Lab|AdobeRGB
...)->deblur
This gives black halos and might be ok from an aesthetic point of view:
Take Photo->convert into linear RGB-> reduce Gamma by
0.5->deblur->increase Gamma by 2.0
Of course, if an image has been blurred in sRGB space, then deblurring
must be done in sRGB space also.
With other words: deblurring must be done in the same colorspace (or in
another linear proportional space) where
the blur was introduced.
That is, why I want to process my images in linear RGB or another linear
colorspace.
Peter
|