That sounds like the approach used in the Pi telecine project. He set the camera to capture a bit outside of the image frame and then somehow compared the position of the current frame to the position of the edge of visible area. He then fed that back to the program to adjust how far the stepper motor should turn.
I'm guessing this is mostly an issue when using the motor to drive the take-up spool directly. If you had the motor moving the film across the gate, then you should be able to make fairly accurate predictions of how much to move the film, since the motor only needs to turn the amount of each frame. In that approach, though, I'm assuming you'd need separate motors to drive the take up and feed reels.