Veritas et Visus September 2008 Vol 3 No 9/10. UCI, p9 FUJIFILM, p7 Pixelux, p23 Eldim, p57

Transcription

1 3rd Dimension Veritas et Visus September 2008 Vol 3 No 9/10 UCI, p9 FUJIFILM, p7 Pixelux, p23 Eldim, p57 Letter from the publisher : Big Sky Country by Mark Fihn 2 News from around the world 4 3DTV Conference 2008, May 28-30, 2008, Istanbul, Turkey 25 SID Display Week 2008, May, Los Angeles, California 34 Stereoscopic Displays and Applications Conference, January 28-30, San Jose, California 46 Behold a Miracle : the rise and slide of the Nimslo camera by Michael Mullen 51 Characterization of autostereoscopic 3D displays by Pierre Boher 57 Eyes tire if focus cues are wrong by Adrian Travis 72 Motion parallax and stereoscopic displays by Robert Patterson 74 Interview with Andrew Fear from nvidia 76 The right horse by Neil Schneider 78 3D game review by Eli Fihn 80 Ray directions - where ray tracing is going by Fluppeteer 81 3D photography commentaries by Sheldon Aronowitz 86 3D HDTV watch by Andrew Woods 88 Fly me to the Moon: a 3D evaluation by Daniel Smith 90 Classification of plano-stereoscopic displays by Lenny Lipton 94 Calendar of events 97 The 3rd Dimension is focused on bringing news and commentary about developments and trends related to the use of 3D displays and supportive components and software. The 3rd Dimension is published electronically 10 times annually by Veritas et Visus, 3305 Chelsea Place, Temple, Texas, USA, Phone: Publisher & Editor-in-Chief Mark Fihn mark@veritasetvisus.com Managing Editor Phillip Hill phill@veritasetvisus.com Contributors Sheldon Aronowitz, Pierre Boher, Eli Fihn, Fluppeteer, Lenny Lipton, Michael Mullen, Robert Patterson, Neil Schneider, Daniel Smith, Adrian Travis, Andrew Woods Subscription rate: US$47.99 annually. Single issues are available for US$7.99 each. Hard copy subscriptions are available upon request, at a rate based on location and mailing method. Copyright 2008 by Veritas et Visus. All rights reserved. Veritas et Visus disclaims any proprietary interest in the marks or names of others.

2 Characterization of autostereoscopic 3D displays using Fourier optics viewing angle instrument Pierre Boher earned an Engineer Degree at the ECP (Ecole Centrale des Arts et Manufactures) in After earning his Ph.D. in material sciences in 1984, he worked in the French Philips Laboratories during nine years on the deposition and characterization of very thin films and multilayers. R&D manager at SOPRA between 1995 and 2002, he was involved in the development of different metrology tools for non-destructive characterization mainly for microelectronics. He joined ELDIM in 2003 to work in the research and development of new metrology heads. by Pierre Boher Optical characterization of autostereoscopic 3D display is mandatory to optimize performances and to make efficient comparison between them. Up to now, quite simple optical characterizations are found in the literature as simple detector moved horizontally at the optimal viewing distance 1. Most sophisticated techniques like goniometers or Fourier optics instruments have also been used but generally in a quite restrict way analyzing only one single cross section in the observer space 2-5. From this limited information, different parameters have been defined such as 3D crosstalk, optimum viewing distance, viewing freedom that gives a first evaluation of the performances of a 3D display 4. The case of multi-view displays is much complex even if the measurement procedures are very similar. In particular, since three or more views can be contributing to the image seen by one eye, the perceived image quality via distribution of crosstalk becomes difficult to analyze even if different attempts have been made in this sense 5. In principle all the 3D display optical characteristics can be derived from the measurement of the output luminance and colors versus viewing angle. Nevertheless an excellent angular resolution is a priori required regarding the high angular resolution of the human eye and a measurement of the entire angular aperture is needed if a full characterization is wanted. Fourier optics instrument are extremely well adapted for this task since the full angle of view is measured rapidly and easily. Nevertheless, the angular resolution of standard instruments is limited to about 0.5, which is not sufficient for accurate 3D display characterization. The purpose of the present paper is twofold; we first present the new ELDIM Fourier optics instrument dedicated to 3D display characterization and its performance; then we propose a new way to analyze the viewing angle measurements on twin view and multi-view 3D displays to get a full characterization of their optical performances. We show that best using conditions of a given display can be visualized easily in the observer space and that an optimum viewing region can be defined and quantified. From this optimum viewing positions all other needed parameters can be computed such as viewing freedom, total luminance, standard contrast or color shifts. Making viewing angle measurements at different locations on the display, 3D spatial corrections can be quantified and verified. The display aspect for each eye of the observer can be also predicted. All these tools offer new precise characterization of autostereoscopic 3D displays for better optimization and easier performance comparison. Viewing angle measurements and Fourier optics Fourier optics viewing angle instruments: Fourier optics has the capacity to transform the angular response of a sample in spatial information that can be imaged by a 2D sensor (cf. figure 1). Each light beam emitted from the sample surface with an angle θ with regards to the normal of the surface is focused on the Fourier plane at the same azimuth and at a position x = F tan(θ). The angular emission of the sample is then measured simply and quickly without any mechanical movement. In practice, the Fourier optics is an achromatic combination of different lenses (6 to 9) that collect quasi all the light coming from the display and focus each angle on an intermediate Fourier plane (cf. figure 2). A field lens and an 57

3 imaging lens are then used to re-image the first Fourier plane on the CCD sensor. In the optical mounting patented by ELDIM, there is field iris before the sensor is complex conjugate of the display surface and allows adjusting the measurement spot size independently than the angular aperture. The consequence is that as represented in figure 2, the measurement spot size varies with the angle. It is what we call cosine compensation which is mandatory to get good collection efficiency even for large incidence angles. The size of the measuring spot can be easily adapted by an iris. Results are generally reported in the Fourier plane where each point corresponds to one incidence angle θ and one azimuth angle φ. The optical resolution mainly depends on the optical setup, the measuring spot size and the pixel size on the CCD detector. It is less than 0.5 for standard ELDIM system. Figure 1 on the left shows a concept of Fourier optics; Figure 2 on the right is the patented optical setup of ELDIM viewing angle instruments. Description of the VCMaster-3D viewing angle instrument: For 3D display characterization, we need to be able to measure all the views of the display independently and with a maximum of angular resolution. The spot size must then be sufficiently large to include tens of pixels for each view of the display. For VCMaster-3D a maximum spot diameter of 4mm has been selected. The optimum working distance is fixed to 20mm. This distance ensures only that the measured spot is always centered at the same location for each incidence angle. It is not critical parameters for the measurements. The main feature of the new instrument is the very high angular resolution below The instrument can measurement full luminance and color viewing angle emission up to 50 using 5 color filters adjusted to the spectral response of the CCD. Main system characteristics are summarized in Table I. Figure 3 on the left is the VCMaster-3D for 3D display characterization; Table I on the right shows the main characteristics of the VCMaster-3D Measurement procedure: For a given stereoscopic display, luminance viewing angle measurements are made for ON state and OFF state and for white applied on each view with black on all the others. The example of a twin view cell phone display is reported in figure 4. The series of measurements is made at the center of the display and 58

4 optionally on other locations. In case of color shifts color measurements can be made instead of luminance measurements. Red, green, and blue states can be also applied instead of white state for more precise color analysis. All the examples provided below are based on contrast between black and white states. In practice it can be also interest to consider contrast one low gray level and one high gray level. It should be closer to the practical using conditions but all the definitions and computations are similar. Figure 4: Viewing angle measurements on one location of a lenticular 3D cell phone display. Computation of 3D optical properties Axis definition and observer position: Our target is to calculate the light arriving from one location of the display to an observer located in front of the display. The emissive properties of this particular location have been of course measured by Fourier optics viewing angle instrument as discussed in the previous paragraph. The observer is defined its coordinates in the XYZ referential (cf. figure 5). The origin O is always the display center. The X axis, Y axis and Z axis define the transverse, saggital and coronal planes respectively as schematically reported in figure 6. The observer position is supposed to be the center of its two eyes. The two eyes are always assumed parallel to the X-axis. The interpupillary distance is fixed (generally 6.25cm). For each observer position we can calculate the two polar angle θ and φ using the following formula: 59

5 (xo, yo, zo) are coordinates of one of the eye of the observer in XYZ referential (xe, ye, 0) are coordinates of the emissive point of the display in XYZ referential The orientation of the head of the observer is also a parameter to deduce the polar angle especially at high incidence angles. We have made the calculations in two limit configurations (cf. figure 7): or the observer keeps the eyes always perpendicular to the display surface independently of his location, or he stare at the location of interest on the display. Of course the result is similar near normal incidence. Figure 5 on the left is a system of coordinates used for the observer and one point on the display surface; Figure 6 on the right show calculations presented in the sagittal, transverse and coronal planes. Figure 7: Observer head position gives different angles: we examine two situations with eyes always perpendicular to display surface or eyes always perpendicular to the direction of the observed point. Problem of the angular resolution: Before going into calculation it is important to understand why the angular resolution of the measurement plays a key role for the observer. The angular resolution is the capacity to distinguish two different light beams coming from the same location but with very close directions. To be realistic the accuracy of the calculation in the plane of the observer must be better or of the other of the human eye pupil diameter. We can easily calculate the maximum position uncertainty x versus the angular resolution θ depending on the observer distance D using the equation: 60

6 As shown in figure 8, the angular resolution of 1 to 2 generally found for spectrophotometers is unacceptable even for phone cell displays with short working distances. The angular resolution of the VCMaster-3D system presented previously can cover all the needs and in particular the 3D TV applications were the working distance is always important. Figure 8: Position uncertainty versus angular resolution: the human eye min pupil diameter gives a criterion for the metrology requirements. Computation for twin view 3D displays: a) Optimum viewing region (OVR): The quality of the 3D display for an observer is directly related to his capacity to see clearly the right images with his right and left eyes. In case of thin view displays, we first calculate the two contrasts associated to each eye using the following equations: (θr, φr) and (θl, φl) are the right and left eye positions in polar coordinates calculated as reported above. YWRBL is the luminance for white view on right eye and black view on left eye, YBRWL is the luminance for black view on right eye and white view on left eye, and YBRBL is the luminance for black view on right and on left eye. These two contrasts are nothing less than the inverse of the 3D crosstalk of right and left eyes χr and χl as introduced by Montgomery in We prefer to work with contrasts because there are quantities used every day in the field of standard displays, and there must be maximize for optimum viewing conditions, which is always easier for graphic display. The 3D quality is optimum only when the two previous contrasts are maximized simultaneously. It is why we have decided to combine the two contrasts to get what we call the 3D contrast for the observer given by: 61

7 We take the product and not the sum because a good quality requires a good contrast for left and right eyes simultaneously. The square maintains the dimensionally of the quantity as a contrast which can be compared to the standard contrast of displays. Figure 9: Calculated contrast for the left and right eye of an observer in the transverse plane: the emissive point is at the center of the display (origin), the extension of the plane is fixed to 50cm in the two directions. To illustrate the definition we have calculated this combined contrast for the twin view phone cell display whose viewing angle measurements have been reported in figure 4. The calculation is made on a surface of 50x50cm and the eyes are assumed to stare at the center of the display. The contrasts for left and right eyes in the transversal plane are reported in figure 9. The different beams at different directions are clearly seen and their intersections define the optimum viewing region (OVR) as shown in figure 10. In the transverse plane (cf. figure 10.a), the main OVR is an elongated area. The two additional lateral regions with high 3D contrast correspond to an inverted situation where the right eye sees the image of the left eye and vice versa. In the sagittal plane (cf. figure 10.b), the OVR is an elliptical shape with quite high spatial extension. This is not surprising with lenticular lenses. We have reported two calculations in two coronal planes at 37 and 28cm from the display surface. 37cm corresponds to the distance of the central position of the OVR that we can define as the Optimum Viewing distance (OVD) and 28cm is better optimized for lateral inverted regions. We see that the lateral position of the observer is always extremely critical as waited for twin view display. b) Definition of additional parameters using OVR: A first thing is to define a criterion for a good 3D factor. Tolerance limit for crosstalk have been reported to be around 5-10% [2]. In terms of 3D contrast we can fix a percentage of the maximum value above which we can accept to be in the OVR. In the case of our example we have made some calculations for 90, 95 and 97% as shown in figure 11 and Table II. For each criterion we can calculate the volume of the OVR and define different Optical Viewing Freedom value along the three spatial directions. We then define: Transverse OVF: maximum extension of the OVR along horizontal X axis. Sagittal OVF: maximum extension of the OVR along vertical Y axis. Coronal OVF : maximum extension of the OVR along depth Z axis. These values are reported in Table II for the three criteria that we have selected. In each case the Tranverse OVF is always very restricted. The quality of the display depends not only on the 3D contrast which take care of the 62

8 differences between the two eyes but also on amount of light arriving inside the two eyes of the observer. It is in particular easy to calculate using the same method as previously the mean luminance value for the two eyes in ON state and OFF state and to evaluate the standard contrast. Results in the sagittal plane are reported in figure 12. Without surprise there is no correlation between 3D and 2D properties and the poor quality of the OFF state on this display outside normal incidence reduce the standard contrast. We have evaluated the variation of the standard parameters inside the OVR for the three previous criteria. There are also reported in table II. Strong standard contrast variations are noticed especially for the less restrictive criterion for the OVR. c) Extension to color shifts using color viewing angle measurement: All the computation made above is based on the luminance emitted by the displays. Color shifts are also important source of imperfection Figure 10: Calculated 3D contrast for an observer in the transverse, sagittal and coronal planes: two positions of the coronal plane are calculated at 37cm (central optimized position) and 28cm (lateral inverted positions). for autostereoscopic displays 5. In the case of lenticular barrier displays for example this effect is difficult to reduce. The twin view phone cell display analyzed previously shows this type of effect (cf. figure 13). Viewing angle color measurements reported in the figure show clearly different angular emissions of blue, green, and red components. The approach proposed previously using 3D contrast ratio based on luminance can be extended using X and Z CIE components in addition to the luminance. We first define the X and Z contrast for right and left eyes of the observer using the following equations: It is then straightforward to combine the contrasts for right and left eyes to get a 3D contrast value for each primary component X and Z in addition to Y: 63

9 In practice we calculate three different OVR for each CIE component. The common spatial region can be considered as the best color OVR for the display. We show this type of calculation for the twin view cell phone display in figure 14. The color shift between red X component and green Y component is clearly seen both on transverse and sagittal planes. The maximum of 3D contrast is not the same for the two quantities. It is always higher for the green component, which is generally optimized, but we can understand that the color OVR can be quite reduced by this type of effect. Additional parameters such as OVD and transverse, sagittal and coronal OVF can be deduced for each CIE component. Some examples are reported in Table III. In this case the red component is the most critical. d) Measurements at different locations and 3D display viewpoint corrections: All types of autostereoscopic displays are normally designed with corrected viewpoint to ensure pixels at the edge of the display are seen correctly by the observer. Figure 11: Different shapes of the OVR in the sagittal plane depending on the criterion on the 3D contrast. Figure 12: OFF state, ON state, 3D contrast and standard contrast in the sagittal plane. 64

10 Table II : Main parameters of the twin view cell phone display depending on the criterion for OVR For lenticular lenses the pitch is adapted so that the center of each pair of pixels should be projected to the center of the viewing window 6. Using viewing angle measurements at different locations on the display surface we can calculate different OVR that should be similar if the display is correctly fabricated. In the computation, we need just to take into account the position of the measurement with regards to the display center. On example of such verification is reported in figure 15 for twin view cell phone display previously analyzed. In addition to the central point, two measurements have been made with a horizontal shift of +2cm and -2cm. Figure 13: Color viewing angles measurement of the twin view cell phone display. Table III: Main parameters of the twin view cell phone display for X, Y and Z component: the criterion to defined OVR is fixed at 95%. 65

11 Figure 14: 3D contrast for X and Y components: red component gives reduced OVR. Figure 15: Computed OVR for three positions on the twin-view display, center and + or -2cm in the horizontal direction. 66

12 Computation for multi-view 3D displays 3D contrast: As reported previously, the optimum viewing region of a twin view autostereoscopic display is very limited and so the observer must be located at a very precise location in front of the display to see it properly. In multiview systems the idea is to increase the OVR by generating multiple simultaneous viewing windows of which an observer sees just two at any time. In this case multi-view systems can support more than one observer. The OVR can be defined in the same way as previously taking into account all the views at the same time. We first define the contrast of view i with regards to all the other views by: Yi(θ,φ) is the luminance at location (θ,φ) when white is applied to view i and black on the other views, YB(θ,φ) is the luminance at location (θ,φ) when black is applied to all the views simultaneously, and N is the total number of views of the display. Ci(θ,φ) is related to the inverse of the crosstalk of view i. The factor (N-1) is here to keep the scale of the contrast comparable for each type of display. For an observer located at (θ,φ) in polar coordinates and with its right and left eyes at (θr,φr) and (θl,φl), the quality of the display is good if the contrast seen by each of his eyes is good. In addition he needs to see different images in each eye simultaneously and only one image must be seen clearly by each of his eyes. We have decided to define a 3D contrast related to the views seen by the observer in the following way. First we calculate the 3D contrast when view i is seen by right eye and view j by left eye: This formula defines one limited OVR were the quality of 3D vision is optimum for view i in right eye and view j in left eye. b) Combined 3D contrasts: This OVR is of course restricted in space and the procedure must be repeated for all the views if ones want an overall performance of the display. We have decided to classify the different OVRs in the following way. We first calculate the combine 3D contrast assuming that left and right eyes see two successive views of the display using the following equation: We call it C 1 because we assume a step 1 between right and left eye. One example for parallax barrier 14 views phone cell display is reported in figure 16. We have reported the 3D contrast between view 1 and view 2 in the transverse (top left) and sagittal planes (top right) and the combined 3D contrast C1 in the same planes (bottom left and right). As waited, the OVR defined by view 1 and view 2 is very restrictive in the observer space. It is in fact comparable to the twin view lenticular lens phone cell display presented previously. The extension in the sagittal plane is nevertheless better because parallax barrier are only dependent on the viewing angle of the LCD and not on the focusing properties of the lenticular lenses as previously. The combined 3D contrast C1 defines a much more extended and continuous OVR in particular in the transverse plane. It means that an observer at the OVD can move horizontally without losing the 3D contrast. He just shifts slightly from one view to the other. In fact in the present example there is an overlapping between the different OVRs that ensure this property. From the computation of C1 we can define different parameters related to the combine OVR such as OVD and OVF in the different directions. We can also define an overlapping ratio between the different limited OVRs. In the same way different other combine 3D contrasts can be defined using the following equation: 67

13 Figure 16: 3D contrast of 14 views cell phone display: between view 1 and view 2 (top) and between each view I and view i+1 (bottom). The calculation is reported in transverse (left) and sagittal (right) planes. The meaning of these new combine 3D contrasts is clear. Computations of C1, C2, C3 and C4 in the case of the parallax barrier phone cell display of figure 16 are reported in figure 17. As expected, different new combine OVRs appear when the observer is a little more distant of the display that the OVD for C1 parameter. These new OVRs allow defining new optical viewing distances related to the type of view seen by the observer. Important overlaps exist also between these OVR that can be quantified and optimized. Computation of viewing aspect: The approach presented above is useful to quantify different important parameters such as OVD and OVF for a given 3D display using Fourier optics viewing angle measurements at one or different locations on the display. Nevertheless when the optimum viewing locations have been determined, the quality of the 3D display cannot be restricted to few parameters like 3D contrast as defined above. To judge of the real 3D quality of a display, an overall evaluation of the display aspect must be made. Computation of the display aspect from optimized viewing position: Knowing the OVR it is then possible to simulate the display aspect using the same viewing angle measurements. The method has been published by ELDIM in different papers 3-4. For conventional LCDs we have shown that their viewing angle capacity plays a key role for the display aspect. A homogeneous white image can appear inhomogeneous for the observer only because of geometrical questions related to its own position in front of the display. The practical standard contrast can fall rapidly even if the normal incidence contrast is high. For autostereoscopic 3D displays the problem is little more complex because of the difficulty to control simultaneously the views of the two eyes. In principle, even for small 3D displays it is necessary to change the emissive properties of the display for the center to the two lateral sides to 68

14 ensure good 3D quality. We have seen an example of this view point correction. Using the display aspect simulation software from ELDIM it is very easy to check what will be seen by the left and right eye of an observer at any location in front of a 3D display. In principle the computation is perfectly accurate only if we know completely the emissive properties of the display at any location on its surface (in practice we make viewing angle measurements at different locations on the display and assume that the variation versus position is continuous). Figure 17: 3D contrasts of 14 views cell phone display in the transverse plane: C1 (top left), C2 (top right), C3 (bottom left) and C4 (bottom right) are reported. Even if we have only one measurement at the center, the display aspect simulation is interesting in the sense that it emphasizes the effect of the geometric parameters. Such a simulation is reported in figure 18 for the twin view phone cell display already studied previously. We assume the observer at one optimized position in the OVR (here on axis at 37.2cm from the display), and we calculate the image seen by his left and right eyes for two situations: white image on left eye and black image on right eye and the opposite. As shown in figure 18, we can predict that the geometric angular correction is not necessary for a very small lateral size of less than 2cm. In practice, cell phone displays are 3 to 4cm large and so a view point correction is generally applied as illustrated in figure 15. The interest of display aspect simulation is not only limited to luminance but also to color as shown in figure 19. The computation conditions are similar than for figure 18 but this time the display color is predicted using a viewing angle color measurement instead of a luminance measurement. The effect of color shifts can be emphasized and quantified. 69

15 Figure 18: Simulated aspect of the twin view display from OVD (37.2cm) for right and left eyes and White- Black or Black-White patterns: the simulated luminance is reported for an emissive 5x5cm surface; the emissive properties are assumed similar to the center at each point of the display. Conclusion Autostereoscopic displays always exhibit complex angular emission properties to provide different images in the eyes of the observer in the best using conditions. A complete characterization requires not only the measurement of the luminance and color in particular directions but a full characterization of the viewing cone with high angular resolution. In this context, Fourier optics instruments appear as the most convenient instrument for this task. In this paper we have presented a new system, VCMaster-3D, dedicated to the characterization of 3D displays that exhibit unprecedented angular resolution. We have introduced also the 3D contrast that is useful to locate the optimum viewing region for an observer in front of the display. Different parameters such as optimum viewing distance and viewing freedom can then be calculated and the display aspect from one given location can be predicted for the right and left eyes of the observer. The approach can be extended to displays presenting color shifts using color viewing angle measurements instead of luminance one s. For multi-view displays, the same approach can be applied and the recovery between the different views can be quantified. We hope that these new tools will be helpful to the engineers in charge of the development of such displays, and also that their will offer better quantified comparisons between displays for the customer. 70

16 Figure 19: Simulated aspect of the twin view display from OVD (37.2cm) for right and left eyes and White- Black or Black-White patterns: the simulated color is reported for an emissive 5x5cm surface; the emissive properties are assumed similar to the center at each point of the display. References 1 H. Hong, Autostereoscopic 2D/3D switching display using electric field driven LC lens, 25.3, SID N. Takanashi, S. Uehara, J. Ishii, H. Hayana, H. Asada, Dual lenticular lense based 2D/3D convertible autostereoscopic display, J. of SID, 335, 12, S. Uehara, N. Ikeda, N. Takanashi, M. Iriguchi, M. Sugimoto, T. Matsuzaki, H. Asada, a 470x235 ppi poly-si TFT LCD for high resolution 2D and 3D autostereoscopic displays, J. of SID, 209, 13, T. Jarvenpaa, M. Salmimaa, Optical characterization methods for autostereoscopic 3D displays, Eurodisplay M. Salmimaa, T. Jarvenpaa, Objective evaluation of multi-view autostereoscopic displays, 20.4, SID J. Montgomery, Analysis of the performance of a flat panel display system convertible between 2D and 3D modes, Proc. SPIE, Vol. 4297, Y. Nojiri, Visual comfort/discomfort and visual fatigue caused by stereoscopic HDTV viewing, Proc. SPIE, Vol. 5291, 303, P. Boher, Study of LCD emission features yields clues to display aspect, Display Devices, Fall 05, N 40, 23, P. Boher, V. Gibour, T. Leroux, Can we predict LCD aspect using local viewing angle measurements, IDW05, 2005 Dec 6-9, Takamatsu, Japan 10 D. Sandin, T. Margolis, J. Ge, J. Girado, T. Peterka, T. DeFanti, The VarrierTM autostereoscopic virtual reality display, Proc. ACM SIGGRAPH, 894, N. Holliman, 3D display systems,