In spite of working mostly for server software and server setup using powerful non-graphic command line tools and scripting languages, it is sometimes fun to work with something very graphical. I did talk about Clojure Art, which is fun and creates interesting visual results and helps getting into the phantastic language Clojure. But more than twenty years ago I have discovered GIMP, which is the main image editor on Linux computers. I keep hearing that Photoshop is even a bit better, but it does not work on my computers, so I do not really care too much about it.
To be clear, I am not a professional image editing specialist, I just do it a bit for fun and without the claim of putting in all the knowledge about colors and their visual appearance, the functionality of gimp and image editing in general… I am just experimenting and finding out what looks interesting or good to me and how to work efficiently. Actually it brings together my three interests, programming, photography and bicycle touring, the combination of the latter two being the major source of my input material.
Now you start working with layers and with tools that increase or decrease the brightness, contrast and saturation of an image or of the layer being worked on. Now I would like to explore how certain functions can be brought into this. Either by putting them into my fork of gimp or into a plugin or into a script within gimp or a standalone script or program.
Some things that I find interesting and would like to explore: additional functions for merging layers. The function has the input of n (for example n=2) pixels from the same position and different layers and it produces another pixel. These are the twenty or thirty layer modes, that describe how a layer is seen on top of the next lower visible layer. So two images of the same size or two layers could be merged. It could be as nicely as in Gimp or just desctructively to make things easier. If it is worth anything for anybody else, maybe making it work as the current modes would be a noble goal. But for the moment it is more interesting what to do. A very logical thing to do is taking just the average of the two layers. So for rgb it could be the arithmetic mean of the r, g and b values of the n layers (or images) belonging to the same x-y-position. Now what would alpha values mean? I would think that they are weighting the average and the new alpha value could be the average of the input alpha values. Now we could use geometric, quadratic and cubic means and with some care concerning the 0 even harmonic means. Very funny effects could be created by combining these byte-values with functions like xor.
When working with any functions, it is always annoying that the r-g-b-values are always between 0 and 255 (or something like that). So this can be changed to real numbers, by doing something like
or to the non-negative real numbers by doing something like
Then some functions can be applied to these double values and in the end the inverse function will just be applied and result in a rounded and limited integral value from 0 to 255.
Do we need this? I do not know. But I think it would be fun to be able play around with some functions on one pixel, a vicinity of pixels, the same pixel-position from different layers and the like.
The nice thing is: we can see the result and like it or throw it away. Which function is correct or useful can be discussed and disagreed on. That is more fun than formally proven correctness, assuming of course, that the functions itself are implemented correctly.
I have not looked into the source code of plugins to tell what can reasonably be done without too much effort. But if someone reading this has some ideas, this would be interesting to hear about.
And finally we have one more advantage of GIMP, because it is open source and it is possible to make changes to it.