Finding blank regions in image

This question is somewhat language-agnostic, but my tool of choice happens to be a numpy array.

What I am doing is taking the difference of two images via PIL:

img = ImageChops.difference(img1, img2)

And I want to find the rectangular regions that contain changes from one picture to another. Of course there's the built in .getbbox() method, but if there are two regions with changes it will return a box from one region to another, and if there are only 1 pixel changes in each corner it will return the whole image.

For instance consider the following where o is a non-zero pixel:

______________________
|o            ooo    |
|      oooo   ooo    |
|      o             |
|      o  o          |
|                    |
|     oo       o     |
|    o  o     ooo    |
|     oo     ooooo   |
|             ooo    |
|              o     |
|____________________|

I'd like to get 4x4-tuples containing the bounding boxes for each non-zero region. For the edge case of the

oooo
o
o  o

structure, I'm not terribly worried how that's handled - either getting both sections separately or together, because the bounds of the inverted-L shape will completely overlap the bounds of the single pixel.

I've never done anything this advanced with image processing so I wanted to get some input before I really write anything (and if there are pre-existing methods in the modules I'm already using, I welcome them!).

My psuedocode-ish version goes something like this:

for line in image:
   started = False
   for pixel in line:
      if pixel and not started:
         started = True
         save start coords
      elif started and not pixel:
         started = False
         save end coords (x - 1 of course)

This should give me a list of coordinates, but then I have to determine if the regions are contiguous.