Skip to content

Latest commit

 

History

History
58 lines (37 loc) · 2.64 KB

astrolabe.md

File metadata and controls

58 lines (37 loc) · 2.64 KB
Raw

Using -S to deal with multiple runs

If you run this script on a file that already has transparency, you'll probably want to use the -S option to suppress antialiasing. But why would you want multiple runs of mktrans on the same file? Because one can clear out "lagoons", areas that should have been transparent but the floodfill didn't reach them. You do that by using the -p x,y option to specify which pixel to start floodfilling at.

Lagoon problem

Here's an example image which has already been run through mktrans once. As you can see, it left a black lagoon in the middle of the loop on this astrolabe.

After first run of mktrans

Eroded pixels problem

We can find the coordinates of the lagoon by running display a-transparent.png and clicking the middle mouse button. In this case, 256,64 is one of the black pixels.

If we did the most obvious thing and just ran

mktrans -p 256,64 a-transparent.png

then the output file (a-transparent-transparent.png) would lose some pixels due to rerunning the antialiasing step on the old transparency.

Second run, but without -S

Kludge: Using -S on subsequent runs

Running mktrans -S supresses antialiasing completely.

mktrans -S -p 256,64 a-transparent.png

As you can see this looks much better:

Second run, with -S

But wait... isn't there a better way?

Yes. There are at least two ways this could be better, but there are no plans to implement them.

  1. mktrans could automatically use -S when it detects a file already contains transparency.

  2. More importantly, -S isn't even the right solution. The proper thing to do is to extract and remove the alpha channel from the original image, run mktrans, and then composite the original alpha channel with the new one.

But, both of those improvements are only necessary if people are going to be running mktrans iteratively and the author (hackerb9) is not convinced that people want to do that. The only use case is when people are using the -p option, which is already a pain in the butt to use as it requires firing up a graphical program, like ImageMagick's display, to find the coordinates of lagoons. Wouldn't people just want to use a program like the GNU Image Manipulation Program instead of using mktrans?

If you think these or other improvements should be made, please file a bug with an explanation of your reasoning.