![]() |
![]() |
![]() |
![]() |
![]() |
||||||
![]() |
||||||||||
![]() |
|
|||||||||
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
Composition Electro-stereoscopic images have an enhanced quality, but does the addition of stereopsis signify that such an image is more "realistic" than the planar version of that image? The idea that stereoscopic images are more realistic than their planar counterparts is accepted by many people. One might say that this has become the commonsense point of view. Although the addition of stereopsis to a display adds another depth sense, it is arguable whether this addition makes the display more realistic. I'm content with the idea that there is an enhancement, but that this is merely an addition of another depth cue to the display. Stereopsis adds information in a form that is both sensually pleasing and useful. If you want to say it's more realistic that's your choice, but technology hasn't reached the point where anybody is fooled into mistaking the display for the world itself. As we shall see, a stereoscopic image which has a one-to-one correspondence or is isomorphic with the visual world may be uncomfortable to look at, and of questionable value for scientist, engineer, technician, and artist. To understand why, and how to control the unique variables of stereoscopic composition, we must first understand them. There are several ways in which a stereoscopic image may depart from being isomorphic with the visual world. We'll discuss three of these conditions which involve, in this order, psycho-physics, the psychology of depth perception, and geometrical or optical considerations. The three conditions are: breakdown of accommodation/convergence, screen surround conflicting cues, and orthostereoscopic conditions. Accommodation/Convergence As stated in the previous chapter, when looking at an object in the visual world our eyes focus ( accommodate by changing the shape of the eyes' lenses) and converge (rotate toward or away from each other). Accommodation and convergence are linked by the necessities of the visual world. When you look at an object, your eyes focus on it so it may be seen clearly, and your eyes also converge on it so that the object may be seen singly. The latter process is called fusion . Different sets of muscles control these functions, and the neurological pathways for these two processes are also separate. Nonetheless, we have grown accustomed or habituated to the linked response because of a lifetime of visual experience. |
||||||||||||
When looking at a stereoscopic display, the eyes accommodate on the plane of the screen but converge based on screen parallax. Thus there is a breakdown of the habitually tied-together responses of two separate mechanisms. For some, but not all, people, this is perceived as discomfort. For the first time in their visual experience, the two sets of muscles are used separately and the unexpected effort may be unpleasant. Many individuals, especially children, don't have any problem with the breakdown of accommodation and convergence. In addition, with practice (which involves no conscious effort), many people become accustomed to viewing stereoscopic images. |
||||||||||||
![]() |
||||||||||||
The problem is exacerbated for the small screens (only a foot or two across) viewed at close distances (a foot and a half to five feet) that are characteristic of electro-stereoscopic workstation displays. Large-screen displays, viewed at greater distances, may be perceived with less effort. There is evidence to indicate that the breakdown of accommodation and convergence is less severe in this case. 1 When the eyes are accommodated for distances of many yards the departures from expected convergence may not be troublesome. This conforms to my experience, and I've been amazed at my agreeable response to stereoscopic motion pictures, typically at theme parks, employing gigantic positive parallax. The breakdown of the relationship between accommodation and convergence is one important way in which a plano-stereoscopic display cannot be isomorphic with the visual world. This psycho-physical fact is one that users, software designers, and stereovideographers must keep in mind, because a plano-stereoscopic display can never overcome this artifact. A direct consequence of this is that, as a rule of thumb, one should keep parallax to the minimum required for a satisfactory depth effect. Accommodation and convergence don't provide the mind with depth information. 2 This statement is at odds with the belief of many stereoscopists. In the case of convergence, the evidence that it provides any depth information has not been clearly demonstrated. For accommodation, it may be the perception of blur (rather than muscles changing the shape of the eyes' lenses) that provides depth information. Therefore, the breakdown of accommodation/convergence probably doesn't result in a conflict of depth cues. While it may seem like a fine distinction, the conflict is actually based on a departure from the habituated relationship of two sets of neural pathways and the muscles they control. The accommodation/convergence relationship is habitual and learned; it is not a physiological given. People who are experienced in looking at stereoscopic images may no longer have the usual accommodation/convergence response. This can be a disadvantage, because stereoscopic images so composed may exceed the limits of what is comfortable for untrained people. Bear this in mind and avoid composing images which tax other people's visual systems. It's a good idea to try your images out on people who are not adept. Screen Surround Looking at a television or monitor screen is like looking through a window, and the plane of the display screen is called the stereo window . The vertical and horizontal edges of the screen are called the surround . As discussed earlier, a stereo image can be thought of as existing in two regions: one within the cathode ray tube (CRT space) and one between it and the viewer (viewer space). When composing a stereoscopic image, there's a basic decision to be made: where to position the scene or the object with respect to the screen surface or stereo window (the boundary layer separating CRT space from viewer space). It's best if the surround doesn't cut off a portion of an object with negative parallax. (By convention, as explained, an object with negative parallax exists in viewer space.) If the surround cuts off or touches an object in viewer space, there will be a conflict of depth cues which many people find objectionable. Because this anomaly never occurs in the visual world, we lack the vocabulary to describe it, so people typically call the effect "blurry," or "out of focus." For many people, this conflict of cues will result in the image being pulled back into CRT space, despite the fact that it has negative parallax. 3 |
||||||||||||
![]() |
Here's what's happening: One cue, the cue of interposition, tells you the object must be behind the window since it's cut off by the surround. The other cue, the stereoscopic cue provided by parallax, tells you that the object is in front of the window. Our experience tells us that when we are looking through a window, objects cannot be between the window and us. You can accept the stereo illusion only if the object is not occluded by a portion of the window. Objects touching the horizontal portions of the surround, top and bottom, are less troublesome than objects touching the vertical portions of the surround. Objects can be hung from or cut off by the horizontal portions of the surround and still look all right. |
|||||||||||
In this regard, the vertical portion of the surround leads to problems. That's because of the paradoxical stereoscopic window effect. When looking outdoors through a real window - for example, out the left edge of the window - the right eye sees more of the outdoor image at the edge than the left eye sees. The same kind of experience occurs for the right edge of the window. The difficulty arises in stereoscopic imaging from the following fact: The right eye, when looking at the left edge of the screen for objects with negative parallax, will see less image, and the left eye will see more. This is at odds with what we see when we look through a real window. This departure from our usual experience may be the cause of the exacerbated difficulty associated with the perception of images at the vertical surrounds. There's an exception to this which is especially important for games developers. Objects which are rapidly racing off the screen may have very large negative parallax values. They tend to look just fine. It's when such objects linger at the screen surround that many people have perceptual problems. Orthostereoscopy There are three conditions to be fulfilled for a stereoscopic image to be orthostereoscopic . 4 Such an image would be isomorphic - in terms of the stereoscopic depth cue and perspective considerations - with the visual world. First, images of very distant objects must cause the viewer's eyes to have parallel lens axes. Second, the distance between the perspective viewpoints used to compute or capture the two views must be equal to the viewer's interocular distance. And third, the perspective constraint must be fulfilled. That is, the image must be viewed from a distance ( V ) that is equal to the product of image magnification ( M ) of the object and the focal length ( f ) of the lens. This is written as: V=Mf This last constraint is well known to lens designers and photographers 5 and holds true for all photographically-created images or images which are computer-generated using a camera model, as described in the next two chapters. |
||||||||||||
![]() |
If we attempted to fulfill the first condition for distant image points, their parallax values would be equal to the interpupillary distance, and that's much too large for comfortable viewing. It's true that viewing distant objects in the visual world will produce parallel axes for the eyes' lenses, but the resultant breakdown of accommodation and convergence for a workstation screen will be uncomfortable. If we tried to fulfill the second condition by specifying the distance between perspective views to be equal to the distance between the eyes, for many applications (such as molecular modeling or aerial mapping) the result would be useless. Molecules are very small, requiring a perspective separation computed on the order of the size of atoms, and aerial mapping often requires a stereo base of hundreds or thousands of feet. |
|||||||||||
Moreover, the distance between human eyes, or the interocular, varies between 50 and 75 millimeters. A view computed for one set of eyes wouldn't produce an orthostereoscopic result for most other eyes. If we were to take the third constraint (the perspective constraint) seriously, we'd have to change our distance to the screen for every new object that appeared. Nobody is going to do this. People have learned how to make a mental adjustment for images produced with wide-angle or telephoto lenses. Although people have asked about the orthostereoscopic condition over the years, I can't recall anybody needing to use it - with the exception of specialized photogrammetric and virtual-reality applications. It's interesting to keep in mind if only to see how one is departing from it. Since you now know the ortho constraints, try an experiment. Create an image conforming to these constraints. How does it look? Now change the compositional parameters in accordance with the advice given here. If the image looks good, that's good enough. ZPS and HIT The parallel lens axes algorithm will be described in Chapter 5. In this software approach, the two views are considered to be created by cameras whose lens axes are parallel. The zero parallax setting (ZPS) is produced through horizontal image translation (HIT) of the two perspective views. When the image fields are shifted so that the object, or a portion of the object, is at the ZPS, that portion of the object or scene will appear to be in the plane of the screen. Composing the image this way helps reduce the breakdown of accommodation and convergence. |
||||||||||||
![]() |
Historically the term "convergence" has been applied to the process used to achieve the zero parallax condition for stereo imaging. The use of the term in this context creates confusion with regard to the physiological process of convergence in the visual world - a process requiring rotation of the eyes. There is ambiguity with regard to which process is meant - physiological or stereoscopic convergence. A further complication arises from this usage: While a portion of the image is said to be "converged" when the corresponding points coincide, the term "plane of convergence" has been used for the zero parallax condition which supposedly exists for all image points located on a plane at a given distance (the distance on which the camera is converged). Unfortunately, the locus of such points will not be a plane if rotation is used for convergence. 4 Rather, it will be a complex but continuous surface. (However, a distortion-free plane of zero parallax will exist when using the parallel lens axes algorithm described and explored in the next two chapters.) |
|||||||||||
![]() |
Therefore, a break with prior usage is called for. The eyes converge on an object in the visual world, or on an image point in a stereoscopic display. Image points in a stereoscopic display appear in the plane of the screen when at the ZPS which, for a computer-generated image using the algorithm described here, is achieved through HIT . |
|||||||||||
We have the ability to control the location of the image in CRT or viewer space, or at the plane of the screen, through setting the HIT. For example, one can put the entire scene or object in front of the window, but it will probably be difficult to look at because of the breakdown of accommodation and convergence, or because of the conflict of cues at the screen surround. Much of the time you're better off making the image appear to be seen through a window. This is especially true if you're creating a scenic view or an interior, as we will discuss in a moment. On the other hand, many computer-generated images involve a single object - for example a molecule or a machine part. For this class of compositions try placing the center of the object in the plane of the screen with some of it in viewer space and some of it in CRT space. This tends to reduce the breakdown of accommodation and convergence. Where you place the object is also an aesthetic decision. |
||||||||||||
![]() |
For compositions resembling scenes in the visual world, most of the stereoscopic image should lie within CRT space. Imagine the following southwestern scene: A cactus is in the foreground, a horse is in the middle distance, and at a great distance we can see mesas. The cactus touches the bottom edge of the screen surround. Let's give it zero parallax where it touches. To give it negative parallax would lead to a conflict of cues, because the stereopsis cue would indicate that the image was in front of the screen while the interposition cue would indicate it was behind. (In this particular example a little negative parallax for the cactus, placing it in viewer space, won't hurt because the conflict of cues for the horizontal edge of the surround is relatively benign.) The most distant objects in the scene, the mesas, will have the maximum positive parallax. Everything else in the scene has between zero parallax and the maximum value of positive parallax. |
|||||||||||
Viewer-Space Effects |
||||||||||||
![]() |
Images with negative parallax values should be carefully treated, and a number of empirical rules apply. Such rules can be broken, but it may save time to learn one approach before trying everything under the sun. One of the most pleasant things about designing stereoscopic computer-generated images is that you can observe what you are doing as you are doing it. If there are troubles with your composition, you can correct them and observe how variations lead to heightened or diminished stereoscopic effects, and how changes may affect visual comfort. Most people enjoy off-screen effects, so it's a technique worth mastering. Some people, vulgarians or the young-at-heart, don't consider an image to be truly three-dimensional unless it pokes them in the nose. An important variable to manipulate for strong off-screen effects is not parallax, but rather the perspective cue. It is the signature of the amateur to use huge parallax values to obtain strong viewer-space effects. Stressing perspective is a better approach, and it's accomplished by using a wide-angle view, at a bare minimum 40 degrees horizontal, and by scaling the object to fill the screen. This will be discussed below and more fully in the following two chapters. Viewer Distance The farther the observer is from the screen, the greater is the allowable parallax, since the important entity is not the distance between homologous points but the convergence angle of the eyes required to fuse the points, as discussed on page 12. A given value of screen parallax viewed from a greater distance produces a lower value of retinal disparity. |
|||||||||||
A stereoscopic image looks more three-dimensional or deeper the farther the observer is from the display screen. Distant objects look farther away, and objects in viewer space look closer to the observer. This subject is treated in more detail elsewhere 3,4 but, simply put, the proportional distances of parts of an image remain constant for any viewer distance from the screen. An object that looks like it's one foot from you if you're two feet from the screen will look like it's two feet from you when you move to a position four feet from the screen. Therefore, off-screen effects will be exaggerated for more distant viewers. Images of objects may, for some compositions, have maximum negative parallax larger than that acceptable for positive parallax. Objects can have large values of negative parallax and be cut off by the vertical surround if they're held on the screen for a short period of time. As mentioned earlier, objects which are racing swiftly off or onto the screen can pass through the vertical surround without being difficult to view. Only experiments will tell you what works best. Remember that in this context the vertical edges of the surround produce more of a perceptual conflict than the horizontal edges. Software Tips The perspective cue weights or scales the stereoscopic cue. If there is a strong perspective cue produced by a wide-angle perspective viewpoint, there will be a strong stereoscopic effect. In order to maintain the image size, the object must be brought closer to the camera when using a wide-angle view. By this means perspective will be stressed. Given a strong stereoscopic effect, the values of parallax may be lowered by reducing the interaxial distance. The stereopsis depth cue may be reduced in this case because the extrastereoscopic depth cue of perspective has been strengthened. The reduction in parallax will diminish the breakdown of convergence and accommodation and ghosting, thereby producing an image more pleasing to view. Here's more advice for software developers who are dealing with images of single objects centered in the middle of the screen: When an image initially appears on the screen, let's allow image points at the center of the object to default to the ZPS condition. Moreover, let's maintain the ZPS within the object even during zooming or changing object size. Some people don't like this notion. They think that the ZPS ought to be in front of the object when it is far away (small) and behind the object when it is close (big), to simulate what they suppose happens in the visual world. But in the visual world the eyes remain converged on the object as it changes distance; hence, retinal disparity, the counterpart of screen parallax, remains constant. The depth cue of relative size is sufficient to make the object appear to be near or far, or changing distance, and often we're only trying to scale the image to make it a convenient size. Its center ought to remain at the plane of the screen for easiest viewing. One thing that's fun about stereoscopic displays has nothing to do with the depth effect. Rather it has to do with glitter, sparkle, and luster. These effects are related to the fact that certain objects refract or reflect light differently to each eye. For example, looking at a diamond, you may see highlight with one eye but not the other. Such is the nature of glitter. This is a fun effect that can be added to a stereoscopic image. It also is an effect that can be used for fire to make it look alive. For example, if you use bright orange for one view and red for the other, you will create the effect of scintillation. For controlling HIT bear in mind that the image fields approach each other until ZPS is reached and then pull away from each other after passing through ZPS. Thus, which way the fields will move relative to each other is dependent upon where the fields are with respect to the ZPS. This parameter is best controlled interactively without using eyewear because the relative position of the two fields is most easily observed by looking at the "superimposed" left and right images. A thoughtful touch is to maintain the ZPS when the interaxial is varied. In this way, the user can concentrate on the strength of the stereoscopic effect without having to be involved in monitoring the ZPS and working the HIT keys. Summary Stereoscopic composition for electronic displays has been discussed by taking into consideration facts about the perception of depth. We have seen that physiology and psychology play an important part in guiding us toward aesthetically-pleasing solutions to the challenges of electro-stereoscopic composition. This in turn controls the design of software. I've stated an opinion that the addition of stereopsis to a display does not necessarily make it more realistic. The addition of stereopsis to a visual display is, in my view, the addition of useful and potentially pleasurable information. Whether or not such a display is more realistic than its planar counterpart is for you to decide. Here are some points to remember, based on what's appeared (and will appear) in this and other chapters in this handbook: Points to Remember References 1. Inoue, T., and Ohzu, H. Measurement of the human factors of 3-D images on a large screen. Large-Screen Projection Displays II , SPIE Vol.1255, 1990. 2. Kaufman, Lloyd. Sight and mind: An introduction to visual perception . New York: Oxford University Press, 1974. 3. Lipton, Lenny. Foundations of the stereoscopic cinema . New York: Van Nostrand Reinhold, 1982. 4. Spottiswoode, Raymond, and Spottiswoode, Nigel. The theory of stereoscopic transmission and its application to the motion picture . Berkeley: University of California Press, 1953. 5. Lipton, Lenny. Independent filmmaking . New York: Simon & Schuster, 1983. |
||||||||||||
![]() |
![]() |
![]() |
![]() |
All materials © Copyright 1996-97, StereoGraphics Corporation |
|||
[Chapter 1] [Chapter 2] [Chapter 3] [Chapter 4] [Chapter 5] [Glossary] | |||
[StereoGraphics] [Products] [How To Buy] [Software] [Developers] [News] [Stereo Cafe] [Support] [Contact Us] | |||