> > A 3 megapixel dslr for $300. > Just take the fall-out sensors > and software skip half of the cells.
An interesting concept, although I'm pretty sure that most of the 'failing' sensors are already sold as 6MP units, and the camera is simply replaced if the end-user complains. You couldn't just skip all the odd-numbered red cells, say; if you had more than two failing red sensors the chances are that you'd have failures on your grid pattern. So you'd have to know where the failing pixels are. Just that simple step of testing and calibrating each individual unit could make these more expensive than the 6MP sensor. You wouldn't want to subsample on a regular pattern, anyway. Take a genuine 6MP image from a scanner (or an image from a digital camera that has been downsized to half the resolution to get rid of any artifacts from the R/G/B interpolation step) and produce two lower-resolution versions of it; one by just taking alternate pixels, and one by using an area-averaging technique. You'll see a considerable difference in quality and artifacts, especially in areas with a lot of fine detail. When you take all that into account, and write software that can (given a list of hot pixels) take appropriate action to skip over them in the image-capture process, why on earth would you sell it as a 3 megapixel camera? You'd be unlikely to have more than 50 hot pixels at the outside (probably a vast over-estimate). Why not just fix them up by interpolating from their neighbours, and sell it as an error-correcting 6MP camera? That's one of the things I like about dcraw; it can be handed a list of hot pixels, and it will interpolate them away. I guess I need to come up with a little tool that can scan a folder full of images and look for possible hot/cold pixels.
