I would go with mocha.
But I can think of one other way then mocha to do this I call it “the cheap way”.
If I got it right, You should be able to do the following:
Start with:
The footage with no actor. Make a freeze frame.
Continue with:
Track the footage with the actor. Set the tracking to the footage with no actor (the freeze frame)
After that:
Mask the actor (guess You have to animate the mask to get it to look good).
When that is done:
Subtract the mask so You can see the freeze frame.
The goal:
By masking the actor away You should see the freeze frame, the freeze frame should have the same shakes as the footage with the actor.
Then You can:
For example make it look better by pre-compose and duplicate the composition and add some gaussian blur and lower the opacity. Or something.. I’m no pro.