[magick-users] Autolevel Correction

[Fred Weinhaus]

Wolfgang,


>Fred's basic idea is to split the three channels R,G,B, adjust each of
>them separately and combine them again to a color photograph. As the
>above example shows, this works just fine. (Anthony: you may use the
>above photographs as an example, as the licence plates cannot be seen.)
>
>The gamm-values mentioned above are just abitrary examples. In the
>skript, I calculate them according to Fred's formula:
>
>gamma = ln(mean_value)/ln(0.5)
>
>This garantuees that in the corrected picture, the mean value is about
>half of the maximum pixel value (i.e. quantum range).
>
>A few remarks however:
>
>1) It is not really clear to me what the difference between
>"-contrast-stretch" and "-linear-stretch" exactly is.


I did a bunch of experiments and Anthony and I exchanged email about 
this quite some time ago. Basically:

I believe you are correct in your explanation, but I would phrase it 
as follows. The function -linear-stretch works by finding the bin 
graylevels corresponding to the clip counts measured from 0 and 
stretches from those graylevels. Whereas, the function 
-contrast-stretch finds the bin graylevels corresponding to the clip 
counts from the min/max bin and stretches from those bin graylevels. 
In either case, counts can be raw cumulative or percent cumulative. 
(This wording can be improved, but this is a quick note for tonight)

Also -contrast-stretch is sensitive to -channel, but -linear-stretch is not.

>
>2) @ Fred: The mean value which is needed for the gamma formula can be
>computed without using the "quantumrange" value, just by
>identify photo.jpg -format "%[fx:mean]"


True, but then it is sensitive to the mean of the image and not 
driven towards a fixed mid-gray value. Could be another option.

>
>3) @ Fred: I think that "-contrast-stretch" does a better job on this as
>your computing of the "-level" values in your script. As far as I
>understood, "-contrast-stretch" actually lets you specify what
>percentage of the pixels will turn to pure black and pure white. (Correct ?)


My script allows you to specify the stretch values if you don't want 
the defaults I chose. It is exactly like -contrast-stretch on each 
channel (or combined channels depending upon what option you choose), 
plus the -gamma is added.

>
>4) The gamma adjustment according to the above formula could be easily
>implemented in IM itself, for instance if the user supplies a gamma
>value of zero, or by some "auto-correct" or "middle" option.



Yes, I already suggested that. But Anthony's and Magick's time is 
limited. Furthermore, Anthony and I are still debating how best to 
improve -contrast-stretch and -linear-stretch.

>Or is there
>any chance that I can get the result of the "identify" statement into
>just one "convert" command, so that I could do the above transformation
>in a simple batch file, instead of having to deal with a tedious VBS?


Not sure what you mean here. But there has been some improvements on 
Anthony's Windows notes. See 
http://www.imagemagick.org/Usage/api/#windows


Fred