Due to the fact that you are seeing the faces in your peripheral vision for a brief moment, you don't have time to absorb all the detail of the face so your brain fills in the gaps based on memory. Your brain takes shortcuts so the results of 'filling in the gaps' are far from accurate.
That's why you don't even need the second image for this effect. The effect goes away if you focus on the actual image instead (using the mouse pointer as reference point)
