Here are two ideas I thought up since yesterday:

1. Raster Keyframer
Would ask users to identify the same objects in all different frames, and would transform the polygon throughout the different frames.

2. Smart Resizer
Create an image matrix any size and space out the dots, kinda like those old 3d views where there were no joined surfaces, only dots, like in star wars. Then it would use a square gradient with four points at the four corners to make the best looking image possible.