A little thread about a new feature for the command line image processing tool I'm working on, Super Irudi. The summary of the new features is this: at the left you have the original picture, and at the right final one.
As the first picture is a pure black & white one, it has only a single channel, therefore, using the "match histogram" feature from Super Irudi results in the following:
$ ./super_irudi.py mars.jpg -i -mh desert.jpg
The picture looks better than the pure B&W one but, still, I won't call it a color picture and I don't really like it. What is the solution I added support for? Well, the problem is colorizing a B&W image, right? Then, why not use an AI like DeOldify, which is Open Source?
And so I did. And, now that Super Irudi supports DeOldify APIs, let's try to colorize my Mars image with a local instance of DeOldify¹:
$ ./super_irudi.py mars.jpg -i --deoldify=http://localhost:5000/process
__
¹ Check this for setting up DeOldify https://github.com/jantic/DeOldify
This picture looks better but isn't what I want: matching the color histogram of my Mars picture with the one from that Earth's desert. But now that we have a picture with 3 channels, matching the histogram is easy: Just append again "-mh desert.jpg" in the command line.
Finally, it looks much-much better! Here you have the 3 pictures:
* The final one.
* The original B&W Mars picture.
* The Earth's desert picture.
The code that supports doing everything I mention in this thread is already available on Github: https://github.com/joxeankoret/super-irudi/
PS: I have also added support for automatically correcting the white balance, by using the command line -ab or --auto-white-balance, but the 'coolest' change, I think, is this one.
@joxean I would be totally disheartened to learn Area 51 is but Soundstage 51 for open air Space scenes......
@joxean this is pretty amaze!
The main idea was to, "simply", match the colors from the panoramic Mars picture I made by stitching lots of little pictures (at the left), with a picture of a desert with mountains I found on the internet (on Pixabay, with permission to use it):