Or alternatively, what's a way to transform high-res images to look like that?
Kinda difficult to buy crappy low resolution detectors these days because nobody makes them and you'd need to build the readout electronics yourself. You might as well just use a decent detector and make it look bad.
I wonder, this may be silly, but could you not potentially get a slightly higher resolution image, if you just created a pixel for every sensel. Than if you were to apply the demosaicing algorithm then convert to grayscale.
That's along the lines of what the Leica Monochrom promises. One drawback of doing it with a regular Bayer sensor is that the colour filters eat light, so while the number of pixels will increase significantly, the spatial resolution may not be that much better.
I just had a little play with the raw data from the Pi camera yesterday using the excellent PiCamera lib.
The performance of the Leica Monochrom is due to this - you don't get more pixels per sensor, but you get proper intensity values at each pixel. This corresponds to better effective spatial resolution in most cases. All they're doing is selling you an identical sensor and pretending it's something fancy and new...
When you buy industrial cameras, the colour and mono versions are identical, but one is filtered (they use the same detector). We use mono in industry often because image quality tends to be much better than colour, and you have a third of the bandwidth requirements (or you want to filter a specific waveband).
Like so: http://imgur.com/a/YpHzZ
https://en.wikipedia.org/wiki/Dither#Algorithms
Floyd-Steinberg is a reasonable first choice if you want to implement one. There's pseudocode on its Wikipedia page too: https://en.wikipedia.org/wiki/Floyd%E2%80%93Steinberg_dither...