Will downscale an image to 1/3 of original height and width, and outputs standard average-based result, and horizontal RGB-order sub-pixel averaged result. If someone cares to make an issue I'll write other sub-pixel types.
If you need a different scale, use another program to scale to 3 times the desired height and width, and then scale it in this program from there.
Takes textual .ppm files only. Use ImageMagick convert example.jpg -compress none example.ppm or similar to get the PPM file, and then e.g.
convert example_spr.ppm example_spr.png to get the result (avoid storing the
result in a lossy format - the tradeoffs of lossy formats are not necessarily
designed to handle sub-pixel concerns, YMMV).
A makefile has been provided to do this kind of conversion, try e.g.
make example_spr.png to output the SP-downscale of
example.jpg, etc...
A comparison image, e.g. example_avg.ppm is also output by the program, for
your convenience. The results are generally very good for photos of natural
environments or people.
3 times the horizontal resolution compared to the standard equivalent. This is a well-known technique for increasing horizontal resolution, I wrote this because I couldn't find an implementation and it was easy enough to write.
Some colorblind people have complained about reading SPR text. I think I understand the problem, and the algorithm can probably be slightly modified to allow better reading for people with this issue.
Obviously the technique depends on the sub-pixel layout of the display device, and won't work unless the image is being displayed at 1:1 scale.
A form of rainbow moiré patterns happen with some imagery, especially fences.
Hard pixel edges and pixel-art becomes pretty and colorful (but not in a useful way).
I've noticed when people write about sub-pixel rendering they're always trying to defend their choice. They will say they're working on a cheaper device, or portable device, or something. Similar things happen when you see research on dithering etc. Please stop doing this, increasing the horizontal resolution of your device by 3 times shouldn't need justification, I'd do it on a 4K phone screen or a 800x600 CRT. Likewise with dithering, we have reasonable parallelizable dithering algorithms, can we just use those on 24-bit downsampled color, instead of building expensive custom 10-bit monitors?
Also why do we say 10-bit color? We don't call 24-bit color 8-bit color.
Seriously, this would be like Intel developing a technique that halved processor lithography, and then writing in their patent "so yeah I guess this is useful... if you're on a Pentium III! hort hort hort".