It’s an unintended consequence of success: disruption. Real D has demonstrated and is introducing a new version of the ZScreen modulator that is twice as bright as the original product that made possible the current stereoscopic electronic cinema. The ZScreen is an electro-optical modulator that can switch the characteristics of polarized light at video field rate. When used in combination with a Texas Instruments equipped DMD light engine, the result is 144 fields per second projected at the screen–half left-handed and half right-handed circularly polarized. This high field rate is required for eliminating motion and stereoscopic judder.
This doubling of light in the soon-to-be-shipping XL (last quarter of 2008) product allows for a substantial increase in screen size. Based on the numbers I have in front of me, right now Real D screens have a median size of 37 feet wide for ‘scope, with a standard deviation of plus-or-minus 7 feet. That number is going to change with the introduction of the XL device. Right now the biggest screens we’re running are about 50 feet. The largest screen we are going to be able to do with the new light doubler will be in the range of 60 feet or more depending upon screen gain and the projector used.
There is a new class of projectors based on a 0.98-inch diagonal DMD chip set, and those tend not to be quite as bright as the prior generation of 1.2-inch diagonal chips. But for various technical reasons that I’m not going to burden you with now, the net effect may well be (because of the 144 fps rep rate that we require) that there may not be much of a difference in some cases especially for 1.85:1 aspect ration projection. Improving brightness for what is now an n excellent projection system is a great step in the right direction. To make it absolutely clear, all things being equal (screen, projector, lamp) the XL ZScreen will be just as bright in stereo as a two projector system using equivalent polarizers. Honestly, I can’t imagine why anybody would want to use a double projector system unless they were in the business of selling projectors.
It’s a great step in the right direction because the stereoscopic selection technology that is used by Real D and our competitors all results in the same kind of light loss. There is a duty cycle loss, because in a field-sequential projection system half the light has to go to one eye and half to the other. So that’s a 50% loss right there. And then the various selection techniques that are used–polarization, or the wavelength selection technique, or the shuttering eyewear that are also employed–all lose about another 70% of the light. The net result is that you are down to about 15% of the light that would have been available.
With a high-gain screen you can boost that. For example, with a screen that has a gain of 2 the net loss would be 70% rather than 85%. By doubling the light we have an extremely beneficial result, and that is achieved by means of polarization recovery. I’m not going to go into the details of how this is done. It’s not the time and place to talk about it. But this technique only applies to the polarization image selection system and is not applicable to the competition systems.
The interesting thing about this improvement–which is a great big improvement because, as I’ve said, the stereoscopic cinema really needs a brighter picture, especially if we’re going to go onto bigger screens–is that the parallax values are now going to magnified even further. Let me explain why this is an issue. It’s understood that images that are composed for one size stereoscopic screen may not be optimum for other sizes. As just a rule of thumb I can tell you based on practical experience that images that are prepared for large screens can be beneficially projected on smaller size screens; but images that are prepared for small screens may have problems on large screens. The reason for this is that the basic stereoscopic depth cue in our displays is parallax. Parallax is the distance between corresponding image points. You could measure parallax with a ruler, but a more sensible way to measure parallax is based on angular parallax, which is based on the distance the observer is from the screen. That is because angular parallax relates directly to retinal disparity. So it’s an invariant, whereas linearly measured parallax is not. (See the blog posting Divergence and Projector Alignment for more background on what is to follow below.)
The reason that this is so important is because exceeding certain parallax values will cause discomfort. There are two major reasons why this comes about. One has to do with divergence. When we look at distant objects in the real world–and that could be objects that are really just tens of yards away–the axes of our left and right eyes are parallel. There is no occasion in the visual field in which they are going to be diverging, or have to verge outward in order to fuse corresponding points on the retina. The analog of parallax on the retina is called disparity, or retinal disparity. So that’s one limit. Based on my experience it’s not a hard and fast limit, and there are some ifs and buts, but it’s the key to what I’m talking about here with regard to understanding this additional screen magnification.
Divergence, in some cases even of large values, is of little consequence to people who are sitting in the middle or the back of the theater. That’s because their eyes are going to diverge a lot less–maybe very little, depending on the size of the screen and where they are sitting. With regard to divergence, the rule is that you’ve got to be kind to the people who are sitting in the front of the theater. But how kind? How much can people tolerate beyond the normal interpupillary separation?
There is an average interpupillary separation of about 64 millimeters for males and females. That means that half of the population is going to have an interpupillary separation that is larger than that, and half is going to have one that is less. Also, based on my experience, you can get away with a lot in terms of divergence. And I’m not even exactly sure what the factors are after all these years of working in the field. It could be that the image in the background is so out of focus, or what’s happening in the foreground is so interesting, that in some shots divergence in the background doesn’t seem to matter or bother people. On the other hand, there are some people who are more sensitive to these things than other people. Another factor that I can’t account for and don’t completely understand is, are we at a point in the stereoscopic cinema where we’re all learning how to look at stereoscopic images, and a little bit of divergence, or maybe even a lot, isn’t going to hurt most people after awhile? The most important factor may have already been mentioned–namely that people who are sitting in the middle or back of the house will have less of an issue fusing divergent issues because the best measure of parallax is angular and not linear.
Most people don’t want to sit in the front few rows of the cinema unless they’re forced to. The people who like to sit in the front rows are special people. I’m not saying that there’s something wrong with them. I’m simply saying that they’re people who just have different taste. I like to sit in the middle of the house. (When I’ve had to sit in the front row because I arrived late, I’ve been pleasantly surprised on many occasions at how modern motion picture projection holds up. I think that’s because the digital intermediate process has eliminated a lot of the contrast grain buildup and sharpness reduction that we had in the past when we used film intermediates.) Be that as it may, people who sit in the front may learn to like the large positive parallax values associated with divergence.
With regard to off screen parallax values, or negative parallax values, there is a lot more latitude. You can take a lot more. People in the field are very sensitive to the concept of the breakdown of accommodation and convergence, which is a habituated response that we have from the time of birth so that the focus of the eyes and the vergence of the eyes are coordinated. In the stereoscopic cinema that is not a major factor despite the fact that it is so frequently cited as a cause for concern. It is well established that for large screen projection–except for the people who might be sitting in the first couple of rows–the breakdown of convergence and accommodation is not a factor. I’m getting tired of people talking about the breakdown of convergence and accommodation.
That being said, you can push parallax values so far that the image is un-fusible. But typically in Real D cinemas (and I can’t give you an absolute number) you can measure off-screen parallax values in many inches–let’s say a foot. But you shouldn’t be measuring parallax values in feet. I can quibble about this, because if you’ve got a spear or something shooting off the screen, in the final frames in which it leaves the screen it could have parallax values of yards, and it will be quite watchable. You might not have time to fuse it, and it will look perfectly fine.
What happens when we go from screens that are maxed out at 50 feet wide to screens that are maxed out at 60 feet or better? Magnification rules, and the parallax values of background points are going to be linearly proportionately larger. In other words, 1 inch of parallax on a 20-foot screen becomes 3 inches of parallax on a 60-foot screen. This is the same for off screen points and for divergent points. It’s the divergent points that are the biggest problem for the especially (and maybe only) people sitting very close to the screen.
So what to do? The advice that we gave to people who were shooting movies for the Real D cinema was this: “Aim for a 40-foot screen. It’s going take of itself.” If you wound up with 2.5 inches of background parallax for a 40 foot screen then on a smaller screen it’s going to be an inch or so and on a larger screen maybe 4 or 4 inches. So nobody’s going to be bothered by the reduction in parallax on the small screen and experience tells me that the image will still look good. For the large screen the major concern is not appearance or stereo effect but background points diverging.
In this discourse up until now I have been more or less concerned with limiting values of parallax in which we’re in a safety zone of comfort of viewing, rather than how the image looks. How the image looks is an interesting subject that needs to be talked about. The eye-brain, or I should say for a stereo system the eyes-brain, is a very flexible instrument. You can enjoy stereoscopic images that were produced for different size screens. As far as I’m concerned, while the image may not be optimum in terms of the extent of the stereoscopic effects, it will be enjoyable. The major issue is not appearance; it’s going to be comfort. That’s why in this article I’ve been more concerned with comfort than I have been with the aesthetics of the image.
I’m going to discuss about a couple of possible approaches. One involves a solution that’s very similar to that which is employed in IMAX. IMAX limits its background parallax points to +2.5 inches and lets everything play off the screen. You can do that in the Real D cinema too by using floating windows. Refer to these blogs and read at the blog that talks about IMAX and Real D compositional differences. This approach requires resetting the background points either in release print or in the theater. I’m suggesting that there could be two skews of release prints, one for very large screens and one for smaller or average screen sizes. The studios will hate this because they, rightfully, want to reduce the number of release print variations. However, the economic advantage to projecting on large screens means you’ve got more people in the audience and more revenue.
Another answer that would be problematical and require a different set of release prints would also involve developing a new product based on cutting edge technology. This would involve interpolating to reduce the interaxial separation. The first suggestion I made would involve horizontal shifts of the image to keep the background points at approximately the interpupillary separation. Conceptually this is easy to do. But once you’re dealing with an image that has been shot with a particular interaxial separation (that is, the distance between the camera lenses), the range of parallax values is baked in. If you want to reduce the parallax range of the film, which might be the ultimate cure for projecting on large screens, the answer could be to interpolate downward to reduce the effect of interaxial separation. This is a much more sophisticated and difficult approach than that which I have suggested with regard to horizontal shifting of the images to maintain a constant zero-divergence background parallax value. It is also a missing link in post production which can vary every important shooting viable except this one.
There is a lot of prior art on interpolation to reduce interaxial separation. Microsoft and others people have been looking into the problem for reasons that are related to applications such as teleconferencing. I’m not going to go into any details about why this has been done, but there is an extensive body of literature on how to interpolate between two perspective views to produce an intermediate a different view.
So there you have it: an issue that needs to be addressed. I’ve tried to lay out the background and give you some ideas on where we can go from here. I’m open to comments and criticism. I look forward to hearing from anybody who’s got better ideas.