The disclosed subject matter relates generally to three-dimensional television (3DTV) technology, and more particularly to a method and system that provide viewers with 3D glasses a 3D experience while viewers without glasses see a 2D image without artifacts such as ghosting. Our approach is applicable to displays using either active-shutter glasses or passive glasses.
With a 3DTV, depth perception is conveyed to the viewer by employing techniques such as stereoscopic display, multi-view display, 2D-plus-depth, or any other form of 3D display. Most modern 3D television sets use an active shutter 3D system or a polarized 3D system and some are autostereoscopic without the need of glasses.
There are several techniques to produce and display 3D moving pictures. A basic requirement for display technologies is to display offset images that are filtered separately to the left and right eye. Two approached have been used to accomplish this: (1) have the viewer wear 3D eyeglasses to filter the separately offset images to each eye, or (2) have the light source split the images directionally into the viewer's eyes, with no 3D glasses required.
As explained in the detailed description below, many 3D displays show 3D images to viewers wearing the special 3D eyeglasses, but show an incomprehensible double image (called “ghosting”) to viewers without glasses. A goal of the present invention is to devise a method and system for providing viewers with glasses a 3D experience while also providing viewers without glasses a 2D image without artifacts.
Many 3D displays show 3D images to viewers wearing special eyeglasses, while showing an incomprehensible double image to viewers without glasses. We demonstrate a method that provides those with eyeglasses a 3D experience while viewers without glasses see a 2D image without artifacts. In addition to separate Left and Right images in each frame, we add a third image, invisible to those with glasses. In the combined view seen by those without glasses, this cancels the Right image, leaving only the Left. If the Left and Right images are of equal brightness, this approach results in low contrast to viewers without glasses. Allowing differential brightness between the Left and Right images improves 2D contrast. We determine that viewers with glasses maintain a strong 3D experience, even when one eye is significantly darker than the other. Since viewers with glasses see a darker image in one eye, they experience a small distortion of perceived depths due to the Pulfrich Effect. This produces illusions similar to those caused by a time delay in one eye. We find that a 40% brightness difference cancels an opposing distortion caused by the typical 8 millisecond delay between the Left and Right images of sequential active-shutter stereoscopic displays. Our technique is applicable to displays using either active-shutter glasses or passive glasses.
Other aspects of the inventive method and system are described below.
We will now describe illustrative embodiments of the present invention. First, we provide an introduction of the problem and our solution, and then a detailed description of illustrative embodiments of the inventive method and system. We will cover methods involving the implementation of a third channel, and brightness of the composite 2D image; experiments and our findings regarding 2D viewer preferences, 3D viewer depth perception, and moving 3D objects and the Pulfrich effect; a prototype system we developed; and a discussion of present limitations and future work to be done. We will also summarize some of the main aspects of our inventive system with reference to
Stereoscopic displays provide a different image to the viewer's right and left eyes to produce a three-dimensional (3D) percept. These displays' falling prices have caused them to grow from a niche product to mass market acceptance with applications in entertainment, medical imaging, and engineering visualization.
The following discussion makes reference to
The most popular 3D display paradigm shows a pair of images on the same screen, intended for the viewers' left and right eyes. The lenses of special shuttered or polarized “stereo glasses” pass images to the correct eye. A viewer not wearing these glasses sees both images superimposed, creating a “ghosted” double-image where two copies of objects appear overlaid (see
We accomplish simultaneous viewing of 3D and 2D images by replacing the pair of images (Left, Right) with a triplet (Left, Right, Neither), where those wearing stereo glasses see the Neither image with neither eye; only those without stereo glasses can see it. The Neither image is the negative of the Right image (see
Unfortunately, this raises the minimum black level of the display for viewers without stereo glasses, drastically decreasing the contrast ratio. This can be mitigated by reducing the brightness of the Right image, αR, to αR≦100%, while maintaining the Left image at full brightness.
If this adjustment is small, the effect on the 3D experience of viewers with stereo glasses is negligible, but the increase in contrast ratio for viewers without glasses is also modest. If this reduction is larger, the improved contrast ratio for viewers without glasses will be significant, but if too large, the 3D experience of viewers with glasses will deteriorate. We conduct experiments identifying the acceptable range of aR for both viewers with and without glasses, and find that both are satisfied when 20%≦αR 69≦60%.
The left and right images may be given unequal brightness either by directly dimming one of the two images, or by adjusting the time allotted to each image, using variable-length frames. We analyze the contrast ratio achieved with each method in Section 3.2 below.
When viewers wearing stereo glasses see a brighter image with one eye than the other, they soon become accustomed to this and report an acceptable 3D experience. However, they also report that horizontally-moving objects appear at different depths than stationary or vertically-moving objects with the same disparity. This small, but measurable, phenomenon is known as the “Pulfrich Effect” and is similar to a time-delay of several milliseconds in their perception of the darker image.
We conduct experiments to quantify this effect. We also measure a similar depth-distortion caused by the 8 millisecond delay between the Left and Right images in a 120 Hz display. The distortion is small enough that it is typically ignored by 3D content creators. We show that these two effects cancel each other when one eye's brightness is 40% that of the other eye. In this regard depth perception is not diminished when one eye is dimmed, but instead is slightly improved.
A primary contribution of this invention is a simple method to allow simultaneous viewing of 3D content by viewers with glasses, and 2D content by viewers without. We support this contribution with experiments measuring: viewer preferences among 2D degradation options, viewer ability to perceive 3D when one eye is dimmed, and the magnitude of the Pulfrich Effect in this system. Lastly, we demonstrate a prototype system built using two commercial 3D projectors.
Didyk et al. have also considered the problem of displaying a 3D image to a viewer wearing glasses while creating an acceptable 2D image for those without glasses, which they refer to as “Backward Compatible Stereo” (Didyk et al. 2011; Didyk et al. 2012). They reduce the disparity between objects in the left and right images to a minimal threshold, preferentially retaining high-frequency components. Smaller disparities make the 2D composite image more acceptable to viewers without glasses, but a ghost image remains.
Anaglyph stereo uses two color channels with passive glasses to provide different views to each eye, while sacrificing color fidelity and showing a double-image to viewers not wearing stereo glasses. The most common example uses red and cyan filters, but amber and blue filters have been used to reduce ghosting seen by viewers not wearing glasses (Sorensen 2004; Ramstad 2011).
Projection on an arbitrary textured object such as a brick wall is possible by adding a color cancelation term to the projected image (Grossberg et al. 2004; Grundhofer and Bimber 2008). We use the same principle, treating one of the stereo channels as a texture to be canceled.
The undesirable ghosting seen by viewers not wearing stereo glasses can also be avoided by using an autostereoscopic 3D display that does not require special glasses. Several techniques have been used to create such displays (Dodgson 2005). For example, a parallax barrier blocks light from reaching proscribed directions (Perlin et al. 2000), and a lenticular array bends light toward the desired direction (Matusik and Pfister 2004). Autostereoscopic displays are generally more complex than glasses-based 3D displays and therefore more expensive.
A typical 3D display produces two images for each frame, (Left, Right). A 3D+2D display produces three images for each frame, (Left, Right, Neither). Stereo glasses ensure that each eye of those wearing glasses sees only one of the three images, while viewers without glasses see the integral of all light. The third field is constructed to cancel one of the standard stereo fields.
A 3D+2D display is not restricted to a single stereo display technology. The key feature required is a third channel of information visible only to those not wearing glasses. In this section, we first discuss implementation options for an additional channel. We then discuss several options for the content of the third channel, which impacts the quality of the composite 2D image.
1. Implementing a Third Channel
Active-shutter displays show each image of the two-image frame packet sequentially, while the lenses of special stereo glasses become transparent or opaque in synchrony to block each eye from seeing images not intended for it. The temporal pattern can easily include more channels, to support our method, or uses such as additional stereo viewpoints (Agrawala et al. 1997) (McDowall et al. 2001).
We believe our method will be most readily adopted by active shutter displays for three reasons. Firstly, it is easily implementable by manufacturers, requiring only a firmware change. Secondly, active-shutter glasses cost as much as ten times more than passive stereo glasses, costing $100 or more, so that owners of active shutter glasses are more hesitant to buy additional pairs of glasses. Thirdly, the Pulfrich effect removes a minor but undesirable depth distortion present in all active-shutter displays.
Two types of passive glasses have been used to create 3D displays. The most common glasses contain polarizing filters of orthogonal polarizations, while the display produces matching polarized images for each eye (Kim and Kim 2005). An alternate option uses lenses with spectral comb filters (Jorke and Fritz 2006). Each eye allows light in a non-overlapping narrow band of red, green, and blue wavelengths to pass. Passive stereo systems may produce the two images simultaneously with a pair of projectors, or on interleaved rows of a flat-screen display, or sequentially with a projector using a rotating filter wheel.
Implementing our method requires three channels instead of two. These could be provided by a single method, such as augmenting the spectral comb filters with a third set of narrow bands, or by combining methods, such as using polarization and spectral comb filters together to produce four orthogonal channels.
Our prototype implementation combines polarization with active shutter projectors to achieve the necessary three channels.
2. Brightness of the Composite 2D Image
Our 3D+2D display shows three images each frame: L, R, and N. Viewers not wearing stereo glasses see only L, because N is chosen as the inverse of R such that N+R yields a uniform grey. The grey field raises the black level of the display: the brightness of the darkest pixel of the screen. If the three images are of equal brightness, the brightest pixel will be only twice as bright as the darkest pixel: a terrible contrast ratio. Allowing the L image to be brighter than the R and N images increases the contrast ratio. Several options are available to produce the N image, with different effects on contrast. We now analyze three possible options, depicted in
Throughout, let L, R, and N be vectors of image pixels, containing all possible brightness values. Let the functions MAX(.) and MIN(.) find the maximum or minimum element in the vector. Let maxL=MAX(L) be the maximum possible brightness for any pixel in L, and similarly define maxR. Let αR=maxR/maxL 190≦100% refer to the brightness of the darker image R relative to L. Let max2D=MAX(L+R+N) be the brightness of the composite 2D image seen by viewers without glasses, and let its darkest possible pixel be min2D=MIN(L+R+N). Since N will be chosen to cancel out R, that is, R+N=maxR, we find that
min2D=MIN(R+N)=maxR=αR·maxL (1)
We now analyze how max2D varies with aR. First, in the simplest Equal-Length implementation of our technique, the three frames (L,R,N) are accorded equal time by the display. In this case, to darken the R and N images, aR is reduced but the brightness of the L image maxL remains unchanged. To cancel R with N, we constrain N to: R+N=maxR, trivially achieved by setting N=maxR−R=αR·maxL 203−R. Since the left frame is allotted one third of the display's photons, maxL=⅓, so the total brightness of the composite image is then:
Second, a Variable-Length display may instead dim the R and N images by affording them a smaller fraction of the total time in comparison to the L image. For example, plasma displays typically form each frame from many shorter microframes, which could be reapportioned unequally among the L, R, and N images. Similarly, some LCD displays now operate internally at very high frame rates of 240, 480, or 960 Hz, interpolating low-frame-rate content. These subframes could easily be deployed to give unequal time to the L image compared to the R and N images. In this case, darkening the R and N images allows a corresponding increase in the brightness of L. In order for N to cancel R, N and R are allotted equal time. Thus, while N=maxR−R, as before, maxL is now constrained as:
We thus find max2D in this case as:
Since αR≦1, the brightest pixel is brighter than in the previous case.
Third, the Equal-Length implementation may be improved. Observe that, as initially described, with αR<100% the N frame never shines with full brightness. Its unused brightness can be repurposed to duplicate L. Before, we had set N=(maxR−R)=(αR·maxL−R). Now, we add L in the unused portion of N:
N=(αR·maxL−R)+(1−αR)·L
With maxL=⅓ as before, the brightness of the composite image in this case is:
This constant brightness falls roughly halfway between the two simpler implementations.
Viewers wearing stereo glasses also experience lower brightness when this system is employed, because precious display time is devoted to emitting photons for viewers without glasses that never reach either eye of those wearing glasses. When using equal length frames, 3D viewers experience a brightness reduction of 33%. With variable length frames the maximum 3D brightness might increase or decrease depending on the choice of αR.
The method introduced in this paper removes ghosting for 2D viewers at the cost of reducing contrast. Contrast can be improved by reducing brightness in one eye for 3D viewers, but this could impact 3D perception. We conducted an experiment to determine viewer preference for contrast versus ghosting, and two experiments to quantify the effect of reduced brightness on 3D perception of stationary objects. Finally, we quantified the Pulfrich effect's depth distortion of horizontally-moving objects.
1. 2D Viewer Preferences
Glasses-based stereoscopic displays superimpose images intended for the left and right eyes on the same space. Viewers wearing stereo glasses see only the appropriate image with each eye, but viewers without glasses see both images with both eyes, producing a double-image with a “ghosted” appearance. Our technique removes this ghosting, but also reduces the contrast. We conducted an experiment to determine at what level of contrast viewers prefer the original, ghosted image to a lower-contrast image without ghosting.
Specifically, in respect to
We conducted experiments simulating simple Equal-Length frames (FIG. 3(top 3a)) and Variable-Length frames (FIG. 3(middle 3b)). Ten subjects participated in each experiment. Subjects' responses were averaged across ten test images, each judged at eleven values of αR. We found that at high contrast levels viewers nearly uniformly prefer our method. Only at very low 2:1 contrast do viewers find contrast reduction equally objectionable as ghosting (
2. 3D Viewer Depth Perception
We display a brighter image to 3D viewers' left eyes than to their right eyes. A modest difference in brightness may be imperceptible, but a completely black right-eye image will obviously preclude stereoscopic vision. We conducted two experiments to quantify depth perception between these two extremes.
We presented subjects, wearing 120 Hz shutter glasses and viewing a 42″ plasma 3DTV, with a stereoscopic display of a 7×3 array of wooden boxes, as in
Six subjects participated in the experiment. Each subject made a total of 130 judgments, across 5 different brightness levels and 13 possible depths.
We find that depth perception is surprisingly robust against differences in image brightness between the two eyes, and is not significantly affected until αR falls below 20% (
In a second experiment, we showed subjects a set of five vertical sticks, as seen in
Viewer ability to perceive depth differences was not impaired until the brightness of the darker eye became very dark, similarly to the previous experiment. Accuracy fell slowly from the equal brightness case until αR fell below 10%. When αR<10%, subjects answered as if guessing randomly.
3. Moving 3D Objects and the Pulfrich Effect
Presenting left and right eyes with unequal brightness has a different effect on moving objects than it does on stationary objects. Accurate depths are reported for stationary objects and those moving vertically, but the depths for objects moving horizontally show a predictable distortion, known as the Pulfrich effect (Pulfrich 1922; Morgan and Thompson 1975). The effect has been used to produce 3D effects in commercial television by distributing tens of millions of paper glasses that feature one dark lens. We conducted an experiment to measure its impact on our system.
We find a small but measurable distortion in the depth of horizontally moving objects, and that this effect can be used to counter another distortion present in sequential-frame 3D displays.
We showed 331 subjects a 3D scene containing two rows of seven stationary boxes, as in the first experiment of section 4.2. The boxes appear to sit at different depths, with the left-most pair of boxes the furthest away, and the right-most pair closest. A moving box repeatedly passed horizontally between the two rows of stationary boxes, as in FIG. (11).
The subjects were asked to identify the stationary box whose depth most closely matched the depth of the moving box. The stationary boxes were unaltered throughout the experiment, but the moving box's speed, direction, and disparity (true depth) were randomly varied. Due to the depth distortions of the Pulfrich Effect, the subjects estimated a consistently different depth for the moving object, depending on its speed and the brightness of each eye.
Three subjects participated in this experiment. Each subject viewed 42 presentations of the scene at each of nine levels of a, with the box moving in a random direction (left or right) at one of seven speeds and at one of three depths (0, 6, or 12 pixels disparity). This experiment was initially conducted with the images shown to both eyes having the same brightness, and was repeated with the left or right eye dimmed relative to the other eye.
Notice that the reported depth error is closest to zero when the left eye is dimmed significantly, rather than when both eyes have equal brightness.
This can be explained by another distortion inherent in shuttered displays. In the upper plot, the blue line corresponds to images of equal brightness shown to both eyes, and is not a horizontal line with no error. This is caused by the sequential nature of active-shutter 3D displays. In a 120 Hz-capable display that shows (left, right) image pairs at 60 Hz, the image shown to the right eye will always lag behind the image shown to the left eye (or vice versa) by 1/120th of a second (8 milliseconds). This time delay causes a speed-dependent distortion in the apparent depth of objects (Dvorak 1872). This depth distortion is often ignored by 3D content creators, e.g. many cameras, the Blu-ray 3D specification, and Nvidia all treat the left and right frames as simultaneous (Vetro et al. 2011) (Gateau and Neuman 2010).
The Pulfrich Effect can be roughly modeled as a time-delay experienced by the dimmer eye. Thus when the left eye is dimmed to approximately 40% the brightness of the right eye, the speed dependent depth-distortion caused by the Pulfrich Effect largely cancels out the distortions caused by the sequential display of left and right stereo images. In the lower plot, we can see that it is not necessary to choose precisely αL=40%. Any darkening of the left eye will lower the error inherent in existing displays.
For consistency of notation and labeling in this application we continue to refer to the right eye as the one that is darkened. However, real implementations on sequential-frame displays should darken the eye presented first.
We have built a prototype of the system using two projectors and a single polarization-preserving screen. The first projector is a standard 3D (120 Hz) projector synced to Nvidia LCD active-shutter glasses and is not polarized. This projector displays the images L and R seen by the left and right eyes of the viewer wearing glasses. The second projector displays the 3rd image, and is linearly polarized. The LCD active-shutter glasses contain an orthogonal linear polarizing element, so that the image from the second projector is not visible.
Note that the first projector spends half its light on the L frame and half on the R frame. The second projector spends all its light on the N frame, but half of this light is lost to the linear polarizer. This leaves all three frames with approximately the same brightness. Geometric and photometric calibration are performed to align the images and correct non-linearities in the projected brightness (Brown et al. 2005).
In the prototype of
We evaluated our system by displaying images in standard 3D, as well as using our 3D+2D method.
Our prototype uses a low quality screen with a significant specular reflection. As a result, our radiometric calibration is only approximate, and ghosting has not been completely eliminated. The screen's preservation of polarization is also imperfect, so that the N image is slightly visible through stereo glasses. Higher quality components and calibration would rectify these issues.
This work has focused on completely eliminating ghosting, and we have compared 2D viewer preferences regarding ghosting and contrast under this assumption. However we have noticed that when the ghost is relatively dim, it is not as objectionable. Further study might reveal an optimum tradeoff between ghosting and contrast reduction by only partially cancelling the ghost image.
In all cases we have analyzed brightness using physical values. However we have observed that when one eye is much brighter than the other, the perceptual brightness is closer to that of the brighter eye. A more careful investigation of perceptual brightness could improve our conclusions about observed brightness when each eye sees an image of different brightness.
Turning now to
Next, a right (“R”) image is displayed for a second period of time, as represented by block 15c. As shown, the “R” image is obtained from storage 15d. As with the “L” image, the display of the “R” image may be coordinated with the toggling of a shutter covering the left eye of the viewer.
The display of the “R” image is followed by the display of the “N” image, for a selected period of time, as shown in block 15e. The “N” image is obtained from storage 15f. As discussed above, this step is designed to substantially cancel out the perception of the “R” image for viewers not wearing 3D eyeglasses, to mitigate ghosting for viewers not wearing 3D eyeglasses. If additional images are to be displayed, as indicated in decision block 15h, the process is repeated. These steps are controlled by a controller 15g, which may be a programmed microprocessor or the like.
As discussed above, the inventive method will likely be more readily adopted by active shutter displays, i.e., since it can be implemented by manufacturers at low cost, allows consumers to avoid purchasing additional pairs of active-shutter glasses, and removes a minor but undesirable depth distortion present in active-shutter displays. Moreover, the inventive method employs three channels (“L”, “R”, “N”), which could be provided by a single method, such as augmenting a pair of spectral comb filters with a third set of narrow bands, or by combining methods, such as using polarization and spectral comb filters together to produce four orthogonal channels. (The prototype implementation described above combines polarization with active shutter projectors to achieve the necessary three channels.) Finally, as explained above, the frame lengths for the “L”, “R” and inverse R frames may be adjusted to optimize the viewers' experience.
3D display technology is quickly growing in popularity. Many current displays require that viewers wishing to see the 3D scene wear special glasses; viewers without glasses not only do not see a 3D scene, but see an unappealing double-image.
We have demonstrated a method to produce 3D displays where viewers wearing glasses see a 3D scene, while those without glasses see a single 2D scene. We have shown that reducing the brightness of one of the images shown to the 3D viewer does not interfere with depth perception, while allowing improved contrast for the 2D viewer. We have also demonstrated that existing depth-distortions in active-shutter displays can be eliminated, due to the Pulfrich effect induced when one eye has reduced brightness.
The true scope of the present invention is not limited to the presently preferred embodiments disclosed herein. For example, the foregoing disclosure of methods and systems for use in making a 3D+2DTV system uses explanatory terms, such as 3DTV, 2DTV, and the like, which should not be construed so as to limit the scope of protection of the following claims, or to otherwise imply that the inventive aspects of the disclosed system are limited to the particular methods and apparatus disclosed. Accordingly, except as they may be expressly so limited, the scope of protection of the following claims is not intended to be limited to the specific embodiments described above.
The following documents are referenced in the foregoing description. These are generally available and copies will be placed in the USPTO file for the present application:
This application claims the benefit of U.S. Provisional Application No. 61/635,075, filed on Apr. 18, 2012, the content of which is hereby incorporated by reference in its entirety.
Filing Document | Filing Date | Country | Kind |
---|---|---|---|
PCT/US2013/032821 | 3/18/2013 | WO | 00 |
Number | Date | Country | |
---|---|---|---|
61635075 | Apr 2012 | US |