Global motion estimation (GME) is a technique that attempts to map one image onto another with a simple four-corner pin. This differs from local motion estimation (LME), which attempts to find where each individual pixel in the image is in the other image. GME is much cheaper to compute than LME, but gives you less information about the image. Nevertheless, it is still very powerful for a variety of applications.
Using the plug-ins’ parameters, you can tell the GME engine what type of motion to expect. This can be a combination of any of:
| 1. | translation - which allows the four corners to translate by the same amount, |
| 2. | rotation - which allows the corners to rotate about their center, |
| 3. | scale - which allows the size of the area defined by the corners to change, |
| 4. | perspective - which allows the angles at the corners to change, so that the area defined by them is no longer a rectangle. |
The more types of motion you allow, the more expensive the motion estimation becomes. For many scenes, rotation and translation are sufficient.
The GME effects have an accuracy control, which controls the amount of work The Foundry’s GME engine does to calculate the global motion. Typically, the higher this is, the better the estimation, but the more expensive it is.
As stated above, global motion estimation simply calculates a four-corner pin to transform one image onto another. This means that GME can’t be used to match two images where there is heavy parallax, very complicated foreground motion, changing objects, and so on.
The best way to think of what GME can do is that if you can do it with a four-corner pin, it can; if you can’t, it can’t. However, GME will take the pain out of hand matching pins frame by frame.