The top image is a crop of the original pixels from a Canon 10D; the bottom is a duplicate which was shrunk to 50% and then expanded back 200% to simulate 25% of the number of capture pixels. After being put in the same image, the composite image was given USM. The top image clearly shows that spatial detail exists at a per-pixel level, and not at half the resolution (25% of the megapixels), as some ignorant people have suggested.