Paul Howson’s Website tdgq.com.au

A Film Restoration Diary

Home-Grown Jitter Stabilisation — Part 1 — Could it be Done?

28 Nov 2016

As mentioned in the previous post, Promoscape had been unable to deal with the jitter: “With regard to stabilisation, we can’t find a cost effective and consistent way to achieve this with the footage.”

They explained that the jitter was not just horizontal and vertical movement (displacement) but included rotation as well.

I’ve talked about possible sources of this jitter in the post Assessing the Scans Part 2 — Jitter. There are a number of possible contributors to the jitter seen in the scans. Here again is the sample of 50 frames from Marianne (1963):

Sample jitter from a corner of a Spirit Datacine scan.
50 frames from Marianne (1963), scanned on the Spirit Datacine at 2048 pixels x 1556 pixels, showing the jitter. What you see here is a square of 70 pixels x 70 pixels from the top right corner of each frame. You can see part of the video raster at the lower left. The white “circle” does not move relative to the video raster. There is one of these little circles in each corner of the frame.

As noted, the partially visible white “circle” at top right remains steady relative to the video raster.

There are in fact four of these partially visible white circles — one in each corner of the film frame. Here you can see the white circles from each of the four corners brought together into a single image:

Sample jitter from four corners of a Spirit Datacine scan.
The same 50 frames, showing a 70 pixel x 70 pixel square sampled from each corner of the frame and brought together into a composite image. Notice that although the white circles more or less move together, there is subtle relative movement, representing rotation and vertical stretch of the frame.

These circles have quite a distinctive shape and a well defined edge. This set me thinking — would it possible to build some custom software to correct the jitter on these kines — software that would take advantage of these clearly defined artefacts in the four corners of each frame, and use them to detect and correct the jitter?

It was worth testing the idea at least.

To do this, I imported a series of consecutive frames into layers in a Photoshop file. Using the bottom layer as a reference, I worked through the other layers one by one, using the Free Transform tool to move, rotate and stretch each layer so that these white circles in each corner precisely lined up with the bottom reference layer.

Having done this, I then exported these layers back into a movie file. Although only nine frames in length, its was clear that the jitter was indeed gone.

Optimistically allowing one minute per frame to do that by hand for the 79675 frames in Marianne would take 220 days working 6 hours per day. A ridiculous idea. But could I create a piece of software that would do it just as well, if not better? That way the computer could do the work, and do it much faster than a human.

In the next post we’ll look at how the jitter — displacement, rotation and stretch — affects the movement of the circles in each corner of the frame.