Wow, this is a pretty novel approach; is this something that's been talked about or are you just one of those really clever people I keep hearing about?

I just learned it from my other question on Stackoverflow not long ago

http://stackoverflow.com/questions/15674880/

It's a cool hack which you can serve low-res, monochrome jpeg in the first scan, and gradually to full-res, full-color jpeg.

Half an interlaced image is still twice as big as an image at half resolution.

Interlacing usually happens in two-dimensions in image formats, if I understand it correctly (ie it first sends one pixel out of each 2x2 block).

The only thing is I think that is one of the actual pixels, rather than an average. In an ideal world you could send the average color of each 2x2 block, then send the 2nd, 3rd, and 4th pixels and use that data to reconstruct the 1st pixel.

Or is that how it works already?