Copyright (C) 2001 Kristian Dupont Knudsen

Fooling The Mind

Motion blur is the effect that arises when a camera captures an image of an object or scenery in motion. The faster the motion, the greater the blur. In an image, motion blur is considered distorting as the blurred parts are harder for the eye to make out. But in a series of frames in a motion picture sequence, motion blur can actually enhance visibility as it levels out the choppiness of the clip. This is the reason why movies look great even though they are recorded at the relatively low frame rate of 24 frames per second.
A common theory goes that the human eye “records” at about 50 frames per second. This has lead to the general misconception that playing any motion picture sequence at 50 frames per second or above will give the spectator the impression of watching “live” material i.e. something that is not a series of separated images. Even if the human eye and brain cannot interpret more than 50 frames per second, it can still tell whether the connection between the frames seems reliable. Thus, images that are motion blurred seem more convincing to the brain than images that are not, if they are watched for 1/50′th of a second a time. Otherwise, they seem less convincing. The best example is probably the rotors of a helicopter. Imagine the rotors rotating at slow speed in the beginning. At this speed, a sharp image might well give a good impression as the difference between each frame is small. But as rotation speed increases, frames with motion blur will seem increasingly more realistic compared to the sharp image that will start to look like random rotation angles displayed in a confusing sequence.

[picture: comic strip with helicopter rotors with and w/o motion blur]

Application

As filmed footage has natural motion blur, only footage that has been created frame-by-frame, such as any kind of animation or computer rendered sequence need motion blur applied. Regular animation is difficult as no information about the lines of motion is stored anywhere, so the animator has to enter it manually somehow. 3D animation, cutout animation and the likes, however, have all the information needed to create a perfect motion blur on all the frames produced.

When looking at products, be it games as well as video editing programs, a lot of them have a confused understanding of what motion blur really is. Or at least what it should be. Motion blur is not the same as motion trails. Many software products offer a “motion blur” that is in fact the last few displayed frames blended together with a low alpha value. This gives a neat distorting effect which may well be the intension for some products (if they merely want to provide a cool effect that makes everything seem a little surreal) but it does not accomplish the goal that can be achieved with real motion blur: an improved visual appearance. This effect compares a bit to motion blur as a Gaussian blur compares to anti aliasing. In fact, you can think of motion blur as anti aliasing over the fourth dimension (considering time to be this dimension), or temporal anti aliasing. Both spatial anti aliasing (in x/y dimensions) and motion blur add information to the displayed data on the behalf of high frequencies. And just like spatial anti aliasing, motion blur does not always improve visual quality. Consider this time line:

This corresponds to the typical behavior of a renderer. Once every 1/25th second (or whatever the frame rate might be), a snapshot is taken of the positions etc. in the scene and this data is used to render an image which is then displayed. The problem here is that all moments represented on the timeline are infinitely small. Compared to a film camera that has an open shutter time of a number of milliseconds for each frame, the timeline would look something like this:

The reason why the motion trails effect is so common is simple: it’s easy to code and it doesn’t require any extra rendering. On the timeline above it is clear though, that if you want to achieve a real motion blur effect, extra rendering needs to take place.

Techniques

Several newer stop motion movies like Chicken Run have artificial motion blur applied by simply taking several shots for each frame, moving the puppets just a little between each shot. This can also be done in a renderer: rendering sub frames and blending them. Ideally, of course, as the shutter-open time is not abrupted, an infinite number of frames should be rendered, or a path should be traced for every pixel creating a mathematically correct blur. I have tried to come up with a way to do so but I find that to be rather complex, especially if you have several objects moving in opposite directions.

Frameless rendering is a way to achieve a kind of temporal super sampling even if it’s not exactly a motion blur. It works by splitting a frame up into pixels, rendering random pixels at various points within the camera open-shutter time. Curious as it may be, I find that from the material I have seen, it does not replace a real motion blur by a long shot. Furthermore, it requires an refresh rate of well over the 25Hz frame rate which might not be a problem on a computer but is if you want to transfer your work to other media such as a television. Finally, the method works best if you have a pixel-by-pixel renderer which is not always the case especially if you render using 3D GPU’s that produce entire polygons (or frames) in one go.

Gabriel J. Brostow and Irfan Essa have come up with an interesting technique that samples two frames and tries to produce an interpolated (motion blurred) result. They’ve developed it with stop motion in mind but it’s definitely worth considering in other contexts as well. It works by dividing the frames into blobs, tiny elements, which they compare in the two frames. This way, they get information about what the motions are and then an interpolation is applied. This technique is probably the closest you can get with no information but the two images whatsoever.

If you have other information, though, you are in luck. In TVA Studio, there are a number of situations where the rendering is done from numbers that have been calculated from a variable time. In such situations, you can render a number of sub-frames. Having not implemented motion blur yet (except from on an experimental basis), I might run into problems that I have not foreseen but I think this is the way to go, maybe rendering a number of sub-frames proportional to the amount of movement in a frame. I believe that Adobe After Effects renders 8 sub-frames regardless of velocities when applying motion blur. This gives fair results most of the time but there are situations (like with the rotors from before) where this is not enough and the result is simply distorted frames that makes visibility much worse.

But Don’t Overdo It..

There is one problem, though. Motion blur tries to trick the brain into believing that something is moving. In real life, however, the eye is capable of focusing on a moving object, reversing the blurring as the brain sees it. This way, the object stands out clear and sharp and the background becomes motion blurred. This is impossible if the blur has already been applied before the image enters the brain. Therefore, you should be careful when selecting objects for motion blur. Objects of interest to the viewer, that remain on the screen for a long time should probably not be motion blurred as the viewer is likely to try to focus on them and failing, the viewers brain will think something is fishy.