A rule of thumb for cameras is that more glass is better, so I’ve always been skeptical of those little bitty chips o’ glass in front of all those pocket-sized camera CCDs.
A while ago I took some shots with my Casio EX-Z850 at various resolutions, chopped out a chunk with some good detail from the middle of the images, and resized (cubic interpolation) the smaller ones to match the 8 MP image. The exposure was automagic: 1/200 sec at f/7.4. Zoomed as far as the optical zoom would go, hand-held ’cause at that shutter speed it’s OK, sharpness and contrast one click higher than the default.
As nearly as I can tell, there’s only a slight difference between 8 & 6 MP, enough to be noticeable at 4 MP, and it’s getting on toward icky at 2 MP. The picture isn’t any bigger because I wanted to preserve the actual dots in the 8 MP image.
Notice the weird color gradations in the shields with bright primary colors. This was hand-held, so the camera reached the same image conversion conclusions each time, starting with different physical pixels. A bit of rummaging turns up the Island Coffee Co logo at http://www.islandcoffee.net/.
Just to show that resolution doesn’t matter in real life, check out the snail. Zoomed all the way out, macro focus, 1/125 f/2.8 (wide open), hand held, resized as above.
As nearly as I can tell, the 2 MP image is the only one that’s noticeably worse and it’s still just fine. The other three are different, but even the 2 MP one isn’t bad.
Moral of the story: lots o’ dots is good, but beyond a few MP it just doesn’t matter. I think 4 MP is the sweet spot, based on an entirely insufficient sample, with 8 MP if you’re planning tight cropping. Otherwise, save 2 MB per image: 8 MP -> 4-5 MB, 4 MP -> 2-3 MB.
The other High Truth: lots of light is a Good Thing. You get really crappy pictures when there’s not enough light, but it seems that’s where all the interesting stuff happens.
Update: Uploaded the PNG version of the shield resolution test, as the JPG was fairly crappy, plus 8 MP overviews of the scenes. The text is antialiased and looks awful when you zoom in, but the whole point is to show the actual dots produced by the camera: there aren’t that many dots for the letters.