1 Centre for Marine Science and Technology,
Curtin University of Technology,
GPO Box U1987, Perth 6845, AUSTRALIA
2 School of Electrical and Computer Engineering,
Curtin University of Technology
GPO Box U1987, Perth 6845, AUSTRALIA
3 School of Physical Sciences, Engineering & Technology,
South Street, Murdoch 6150, AUSTRALIA
Keywords: 3D video, stereoscopic video, field-sequential, standards conversion, PAL, NTSC, SECAM.
Unfortunately, despite its simplicity, field-sequential 3D video cannot be converted between the 60Hz (NTSC) and 50Hz (PAL & SECAM) video standards using conventional standards converters. Most, if not all, video standards converters corrupt the 3D content of the 3D video signal by mixing the odd and even fields and the output signal is unviewable in 3D in the new standard.
The standards differ in three main respects: the field rate (number of fields per second), the number of lines per frame and the method of encoding colour. The parameters for each of the three standards are summarised in Table 1.
|colour encoding||QAM 3.58MHz||QAM PAL 4.43MHz||FM 4.25, 4.40MHz|
All three standards use 2:1 interlacing. This means that each frame of 525 or 625 lines is scanned in two parts called fields. Firstly all the odd numbered lines are scanned (called the odd field) and then the even numbered lines are scanned (called the even field). Therefore there are 262.5 lines per field in the 60Hz standard (NTSC) and 312.5 lines per field in the 50Hz standards (PAL & SECAM). Interlacing is used because it allows a high vertical resolution with a low amount of flicker while keeping the signal bandwidth to a minimum.
Field-sequential 3D video does have a problem with flicker when used with a standard television because each eye only receives half the overall field rate (25Hz for PAL and 30Hz for NTSC). The flicker problem can be overcome by using commercially available field doublers 2.
With field-sequential 3D video there are two polarities by which left and right images can be stored in the odd and even fields. Most companies have chosen to store right images in the odd fields and left images in the even fields (3DTV Corporation, VRex Inc, Virtual I/O, SOCS Research, etc). Some systems, however, use the opposite polarity, e.g. the Toshiba 3D camcorder. The result of this is that 3D video generated with one system cannot be viewed correctly on a system with the opposite polarity. The incorrect image will be sent to each eye and a pseudoscopic (reversed stereo) image will be seen by the viewer and incorrect depth information will be perceived. Some systems are, however, compatible with both polarities by changing an external switch.
The squeezing method has the advantage that allows the 4:3 aspect ratio (ratio of image width to image height) of the left and right images to be retained (after they are unsqueezed). Digital video electronics are used to squeeze the left and right images from a pair of video cameras to generate the sidefield 3D video signal. Digital video electronics are again used at the display to convert the sidefield format signal into a 120Hz field-sequential signal. The sidefield 3D video signal can be recorded and played back with a standard VCR. To our knowledge, the generation of 3D video in this format is only supported by equipment available from StereoGraphics Corporation 3 (San Rafael, California). 3D video in this format can also be displayed on some equipment available from 3DTV corporation (San Rafael, California).
The sidefield method without the 2:1 horizontal image squeeze is generally only used for amateur purposes because the images have a vertically narrow aspect ratio of 2:3. Sidefield 3D video in this format is generally produced by a single video camera fitted with an optical beam splitter. This is basically a device containing four mirrors and is quite commonly used in 3D still photography. The image is viewed in 3D either by free-viewing the stereo-pair or by viewing the display while using some optical aid (containing either mirrors or lenses).
We are only aware of one system which used this format with standard video. It was developed by StereoGraphics Corporation and implemented in the NTSC video standard.4 This system was discontinued several years ago when StereoGraphics' sidefield system was released.3 The use of the subfield format with standard video is not supported by any currently available equipment.
Since this system was invented before inexpensive digital video electronics were available, it required the use of specially modified video cameras to generate the subfield format 3D Video. The signal was displayed on a monitor whose vertical deflection scanned at twice the normal rate so that the left and right images were displayed overlapping each other. The image was then viewed through a pair of shutter glasses which were driven in synchronisation with the left and right images being displayed on the screen. This system had the big advantage that the 3D imagery was displayed flicker-free, however this advantage was offset by the complexity of the cameras. The subfield format 3D video signal can also be recorded and played back with a standard VCR.
Conversion of the colour encoding method is a fairly simple process and can be relatively easily achieved using linear analog electronics. Unfortunately, the process of changing the field rate and the number of lines per frame is more complicated and is generally performed using digital electronics. There are three main ways in which the number of fields per second and the number of lines per field are converted: Field/Line Omission/Duplication, Field/Line Interpolation and Motion Estimation.
This is the simplest and lowest quality conversion technique. It introduces some conversion artefacts especially when motion is present in the scene. Subjectively the conversion is acceptable.
In a simple implementation of such a system, a new line in the output standard is calculated as a linear interpolation between two lines from the input standard. The particular input lines from which the output line is calculated and the weightings used are determined from the position in the scan where the output line must be generated. This is illustrated in Figure 2(a) which shows a PAL to NTSC conversion. For example, line 5 in the output standard is calculated as 24% of line 5 and 76% of line 6 from the input standard. This calculation continues such that the correct number of output lines is generated from the input lines. The conversion of the number of fields per second is a similar process and is illustrated in Figure 2(b). For example, output field number 3 occurs at t=2/60 seconds. It is calculated from inputs field numbers 2 and 3 (which occur at t=1/50 and t=2/50 seconds) at a weighting of 33% of field 2 and 67% of field 3.
Four line/four field converters are also available. They work in a similar way to the process explained above except that each individual output line is based on a weighted average of four input lines and each individual output field is based on the weighted average of four input fields.
The performance of this conversion method with standard video is much better than the previous method, however some conversion artefacts are still evident, particularly with scene motion. It should be noted that the details I have provided give only a brief description of the process. A full explanation is contained in Sandbank (1990).
Motion estimation is generally only found in broadcast quality standards converters. The quality of conversion will obviously vary with the quality of the algorithm which calculates the motion vector array.
Three different types of problems occur with each of the three methods of standards conversion. These problems are illustrated in Figure 3 for a PAL to NTSC conversion. Figure 3(a) show the native PAL field sequential 3D video signal. The black and white squares represent the odd and even fields which contain right and left images. The field/line omission/duplication method (illustrated in Figure 3(b)) does not mix fields, but the field polarity of the output signal changes every five or six fields. It can be seen that where a field is duplicated (the two consecutive white fields or the two consecutive black fields), the field polarity changes. This obviously destroys the 3D effect. The field/line interpolation method corrupts the 3D information because it produces the output fields from a mixture of odd and even fields. As can be seen in Figure 3(c), very rarely in the output field sequence does a complete left image or complete right image exist. Generally the output fields are a mixture of left and right input fields (represented by the different shades of grey). The motion estimation method corrupts the field-sequential 3D video signal because output fields are motion estimated between consecutive odd and even fields and therefore between a pair of right and left images. The corruption of the 3D information would not occur if even output fields were motion estimates from a consecutive pair of even fields from the input standard. Unfortunately, this is not the case with currently available equipment.
The separate channels method of 3D video has slight problems with standards conversion. Care must be taken that a time shift is not generated between the two channels when the conversion takes place. If a time difference is introduced between the two channels, temporal stereoscopic effects would be introduced which could upset the stereoscopic information. Ideally both channels would be converted simultaneously with a pair of standards converters (with synchronised output timebase) and with the input video signal coming from a pair of synchronised VCRs.
The 3D information in the other three 3D video methods (sidefields, subfields and anaglyph) is not corrupted by the standards conversion process. The only conversion artefacts present are the same as those present when converting normal (non-3D) video between standards but this does not corrupt the 3D information.
We have extended the capabilities of a commercially available video standards converter to allow the successful conversion of field-sequential 3D video between the PAL, NTSC and SECAM video standards. The converter allows field-sequential 3D-PAL, 3D-NTSC and 3D-SECAM material to be converted to field-sequential 3D-NTSC or 3D- PAL. Particular care is taken to keep the odd and even fields separate in the conversion process and to ensure that the left and right images from the input standard are stored on the even and odd fields of the output video standard. Figure 4 shows how the converter can be used to convert field-sequential 3D-NTSC to 3D-PAL.
The advantage of using digital video and digital frame store technology in the implementation of a standards converter is that it also allows a number of other functions to be achieved. The converter can be used for (a) the conversion of field-sequential 3D video to 2D and (b) the conversion of field-sequential 3D video to the opposite field polarity (field inversion). In the 3D to 2D conversion mode the output video signal consists of only odd fields (left images) or only even fields (right images) of the original field-sequential 3D video signal as chosen by the user. This mode could be used to convert a 3D video sequence to 2D so that the footage could be shown to an audience without need for 3D viewing apparatus. Another application of this mode is for 3D video projection. If two converters are used along with two video projectors, one converter could be configured to provide the first video projector with left images only and the other converter could be configured to provide the second video projector with right images only. If polarising filters are placed in front of each of the projectors and a silvered projection screen is used, a stereoscopic video projection display would be achieved. This configuration is illustrated in Figure 5.
The field-reversal mode swaps the field polarity of the incoming video signal. Images stored on the odd fields are shifted to the even fields and vice versa. For example, this mode could be used to convert field-sequential 3D video which has been recorded with the Toshiba 3D camcorder (which stores left images in the odd fields) to the defacto standard field polarity (left images stored in the even fields). This configuration is illustrated in Figure 6.
The converter also acts as a time base corrector to stabilise the timing of the video signal and clean up the synchronisation signals.
2. Andrew Woods, Tom Docherty and Rolf Koch, "Field Trials of Stereoscopic Video with an Underwater Remotely Operated Vehicle", Stereoscopic Displays and Applications V, Stereoscopic Displays and Virtual Reality Systems, J. Merritt, S. Fisher, Editors, Proceedings of the SPIE volume 2177, pp. 203-210, 1994.
3. Lenny Lipton, "Stereoscopic Real-Time and Multiplexed Video System", Stereoscopic Displays and Applications IV, J. Merritt, S. Fisher, Editors, Proceedings of the SPIE volume 1915, pp. 6-11, 1993.
4. Lenny Lipton, Lhary Meyer, "A Time-Multiplexed Two Times Vertical Frequency Stereoscopic Video System", 1984 SID International Symposium, Society for Information Display.
5. C.P. Sandbank, "Digital Television", John Wiley and Sons, Ltd., West Sussex, 1990.
The 3D Video Multi-standard Converter mentioned in this article is available for purchase. Click here to see the brochure.
Copyright on this document is retained by Curtin University. This document is not public domain. Permission is hereby given to reprint this paper on a non-profit basis for scholarly purposes provided the document is unaltered and this notice is intact. This paper may not be reprinted for profit or in an anthology without prior written permission. If you wish to reprint this paper on this basis, please contact the primary author at the address shown on the first page of this document.
Last modified: 16th February, 1996.
Maintained by: Andrew Woods