When analyzing an image I'm used to thinking of the pixels as a grid of boxes the photons have fallen into. I assume that there's zero space between the boxes. That seems obviously wrong, since there needs to be some separation on the detector to prevent electron bleed between pixels. So, what's the real geometry of pixels in a detector like (especially for: CCDs and infrared arrays)? Is it closer to a grid of points than a grid of boxes?
Usually the separation exists, but is negligible compared to the size of the pixels. Hence, the loss due to photons "falling through the cracks" is usually very small, and can be well accounted for using either Monte Carlo-based methods or experimentally using a focused source of known brightness. For example, attached is the user manual for the WIYN One-Degree Imager (ODI), a CCD detector that I have done some work on. There is a section on detector geometry that will probably help.