Stabilizing Footage in Paint*
By Jon Carroll

I recently had the opportunity to shoot a friend's college graduation on video. Like most of these sort of things, I had about three days notice and was forced to shoot under less than ideal conditions. Mostly, the big problem here was camera shake. There wasn't space to set up a tripod, so I used one of those lap monopods with the knowledge that some camera shake would be induced by things like me shifting in my seat. But, I knew I would be editing it later, and that I could clean up some of the camera shake by stabilizing it in Paint at the same time.

I'm still using the same setup I was when I wrote my earlier tutorial on Effect*, so I'm accessing my video as Targa frame sequences on my DPS Perception. Unfortunately, because of the length of the ceremony before editing, I couldn't capture the footage at the best quality my PVR supports. Instead, I had to capture at the best quality that would fit on the PVR's dedicated drive.

First I took the main footage and trimmed it into several separate sequences based on the planned edit points. I also trimmed the audio because I won't be changing the length of the sequence, so I can use my editor to attach the audio from the non-stabilized sequence. (This is because the PVR stores its audio and video as separate files, the audio is stored on a system drive while the video is stored on the PVR's dedicated drive. DPS uses a file called an Audio-Video Clip [.AVC] file to store the location of both files so their player utility will play both simultaneously.)

The footage loaded into Paint* I then opened the image sequence in Paint* with and set the appropriate pixel aspect ratio in the  File > Footage Settings Window (PVRs use D1 pixel aspect ratio.) Since the sequence was a bit long, it would take awhile to track each sequence (sigh, the only disadvantage) If I were using Edit* version 5.0, I could simply set up the track and save the project settings then import the clip as a project into Edit*, saving myself the trouble of the image getting recompressed twice (compressed once on capture, again with the stabilized sequence, and again after editing.) But alas, Discreet does not support DPS products directly like they do with Truevision's (now owned by Pinnacle) Targa series, and the Matrox DigiSuite.

The Stabilize 2 points window. Next, I went into the Effects menu, and selected Stabilize >Stabilize 2 points... because I intended to use more than one point in the image to provide stabilization data, and to stabilize for rotation and scaling as well as position. Positioning the first Stabilize point Once again we come to the arcane practice of finding tracking points in the image. Remember, this time the points not only have to have enough contrast and relative stability (i.e. there are no major changes in the tracking point, no one walks in front of it etc.) but they also have to show where the motion is you want to stabilize. In this case, my first tracking point went along the upper edge of the podium right where the microphone crosses the edge. To do this, click on 'pick' next to 'Stabilize point 1' and manually position the stabilization point crosshairs. Then, you will need to turn the tracker on for 'Position' and create a tracking point on the Stabilize point you selected. you can now adjust the tolerance for the tracker for this point, including the size of image it uses for the track. Keep it fairly tight because you don't want the stabilization to wander.

Once you position the tracker, position the Analyze button in the tracker palette and track the footage from this point. Positioning the tracker on the second Stabilize point. Next, I selected a second tracking point. This time, I picked an outer corner of a planter halfway across the platform. The  nice black planter provided a lot of contrast against the platform itself,  and the corner stayed a clear enough structure throughout this part of the shot.

Once I positioned the stabilizing point, I activated the tracker and analyzed the footage just as I did with the first point. After a bit of experimentation, I decided to move the point to the back edge of the table cover, because with the stabilization set to 'fit' mode, the field separation of my video footage became too apparent with the points being so close together vertically.

Also, another thing I noticed is the need to make large tracker boxes. I ended up going back and making the tracker boxes twice the size that you saw above in order to get the fit of the image correct. But, it did stabilize the footage.

Included here are two mpegs, before and after stabilization.
Click the left side of the image for the before clip, and the right
side for the after clip.

Jon Carroll is a freelance animator, illustrator and production designer located in Richmond, VA. He can be reached at draven@infi.net