Just having 20 images is probably easiest. I've used this technique also for compositing different video's together in Processing using different 'layers' and masks etc. But for the sake of brainstorming here are some more ideas...

1. Have one big composite source image (all 20 images next to each other) and use copy, get-set or pixels[].

2. Pre-analyse the frame differences (1->2, 2->3 etc.) per pixel between your frames and only change those pixels that (a) change and (b) are not behind a webcam video... using bitshifting techniques!

Absolutely no idea if these methods would be faster, but it would be nice to find out.