23 March 2011

Displaying images in 3D has had any number of starts in the past 100 years. It gets closer to becoming a long term reality as display technology becomes more sophisticated and more affordable. For the first time in the many attempts at making 3D more than a passing fancy we have a number of ways it can be done. Not only is there a race to make 3D a big part of consumers' future, but there is a competition among several approaches to get there.

One of the early attempts to put 3D into your home has already gone by the wayside. Think colored glasses, the anaglyph approach. It didn't take long for consumers to see through the weaknesses of that system.

We see complications ahead trying to determine the worth of the other approaches. Part of the problem is most people don't know what to expect from 3D. Almost anything they see from any of the remaining approaches looks 'good', partially because consumers have no prior knowledge of what 3D should or could look like. Most of us need enough experience with high quality 3D before we'll be able to get beyond the wow factor of the format.

For the time being, until we get holographic displays, 3D is essentially two 2D images, one for each eye. A 3D display system has to get the left eye image to your left eye and the right eye image to your right eye. There are two basic approaches to making this happen on flat panel displays.

One approach is passive, where the image is polarized one way for one eye and polarized another way for the other eye. Both left and right eye images are on screen at the same time. Polarized glasses are worn by the viewer so that each eye sees the appropriate image. Since the screen has a fixed resolution of 1920 by 1080, the horizontal or vertical resolution of the two images must be cut in half in order for both to fit on the screen at the same time.

At the Consumer Electronics Show in Las Vegas this past January a new type of LCD flat panel was introduced. It's being called Film Patterned Retarder, FPR technology. It's a filter that goes on the front of an LCD panel that polarizes odd lines in the picture one way and even numbered lines the other way. If we take out all of the odd numbered lines out of one of the two full resolution images we can assign it to a polarization for one eye. If we take all of the even numbered lines out of the other image we can assign it to the polarization of the other eye. The lenses in the passive polarized glasses will allow light from one polarization to go to the left eye and the other polarization to the other eye.

What you see, at best, if there is no overlap of the filters on the surface of the LCD panel, is two images, each with potentially half of the vertical resolution of the source image. (We'll discuss the resolution of the source image further on in this article.)

The second approach to 3D uses active glasses and full resolution sequential images. The full resolution left eye image is displayed on the screen while the active glasses allow light from that image to go to the left eye, but not the right eye. In the next period of time the full resolution right image is displayed with the glasses allowing the right eye to see the image, but not the left eye. Our brain retains both images long enough for us to believe both images are always there.

The active glasses approach to 3D has been on the market for some time. It's a natural for most display technologies plasma and LCD in flat panels and DLP in projectors. Sales of such sets in 2010 weren't exactly good. There could be any number of reasons beyond the economy. Among those listed by trade organizations, many of the reasons center around the active glasses themselves. There is the cost of the glasses, their weight, their style, and the fact that they are battery powered and must be periodically charged.

For the time being, maybe always, the LCD panels using the FPR technology are more expensive to make, but the passive glasses are less expensive.

Some interesting claims are being made for the FPR approach to 3D in an attempt to position it against the shuttered glasses approach. It's easy to claim full resolution capability for the shuttered glasses since full resolution left eye and right eye images are being displayed. But the FPR camp is also hinting at a full resolution capability when each of the two images is being displayed at half of the vertical resolution of the source. You know, two halves make a whole. Then again, if the image is truly 3D the two halves are not the same. They don't just add together. None the less claims are being made for essentially no noticeable loss in picture resolution.

At JKP we've heard those claims for some of the transmission systems that also cut vertical or horizontal resolution in half in the process of conveying 3D images to consumers. Here is where we get into source resolution. Pushing full HD resolution left- and right-eye images through a TV channel to create one full-HD 3D image on the display can be extremely difficult. The two images are often put into the space of one 2D image by cutting the vertical or horizontal resolution of each image in half. The two images are then placed side-by-side, or top and bottom.

Imagine a situation where the horizontal resolution is cut in half for transmission, which is often the case, then being displayed by a TV that cuts the vertical resolution in half.

The only 3D delivery system currently claiming full resolution left and right eye images is Blu-ray.

Justification for cutting the source signal in half has come from several studies saying consumers don't see the difference between full resolution and half resolution sources when displayed on full resolution sets. We've been disappointed to hear papers being presented at technical conferences stating this. We know that in ideal viewing conditions that shouldn't be the case, which leaves us wondering how these surveys are being done.

From our experience in demonstrating 2D images we have serious issues with any clams consumers can't see the difference in full resolution versus half resolution images. As an example, when viewing properly demonstrated images, few audience members have ever missed the difference between standard definition composite video and component video. When shown good HD from broadcast sources versus the best that can be done in Blu-ray, no one misses the differences. In fact I've been asked how I stand to watch broadcast or satellite HD when I have so many titles on either HD DVD or Blu-ray.

We all know that consumer impressions can be easily swayed by the way they are being presented. It's even easier to do this in 3D since most viewers have no reference for what it should look like. The question becomes, how do you condudct objective research about what can be seen in 3D imaging?

JKP is advocating that the quality of 3D must be evaluated by looking at 3D as two 2D images, each of which must stand up to the well-established rigor we use to evaluate 2D images. Objections to our position have been voiced saying 3D is more than two 2D images. We agree, there are additional issues beyond 2D, but we argue what we are proposing is a better starting point than anything else being offered. In any case you must get by what the two 2D images are doing before going on to the issues surrounding 3D. We are suggesting that the evaluation process be done one step at a time. Start with what we know well, 2D images, and build on that for a comprehensive approach to evaluating 3D.

JKP can help in defining a critical part of evaluating 3D. Some declarations of our history in the industry need to be made before going further. We are bringing a bias of our past work into this discussion.

Number one, Samsung employed JKP for a number of years as a consultant in image quality. They've come back to us asking if we would look at both the active glasses and passive glasses approach to 3D.

Number two, we at JKP have long supported the full resolution dual 2D image approach to 3D. We believe images can be sequenced fast enough for the viewer to believe both are always there. We mostly believe 3D should not compromise the 2D image, as the majority of program content you'll be watching in the next few years will be 2D.

As we've stated, our approach to evaluating 3D quality starts by looking at the two 2D images that make up the 3D picture. Even more basic, we believe the 2D performance of a combination 2D and 3D display will tell you a lot about what to expect from the 3D side of the set. It is highly unlikely that the dual 2D performance will be better than the single 2D performance. That said, all of our evaluations of 3D sets start by looking at their 2D capability.

We also feel that possible compromises in bringing 3D signals to your home should be taken out of the evaluation process of the display. The best possible, full bandwidth signals should be used as a signal source. We generate many of our own test signals to insure they represent the best we can deliver.

We have a new ally in this effort. AV Foundry has created a generator, the Video Forge, which can accept our test patterns and provide them to the TV set in the frame sequential format. Our test patterns are generated at full 1080p image resolution in the RGB domain. They are compressed to the lossless png format for delivery to the generator. The generator in turn provides the set with full bandwidth, uncompressed RGB signals to the display. The generator will allow us to send full bandwidth images for both 2D and 3D to the TV.

This means our testing can be done with the best possible source material. It is full resolution, with no losses due to conventional distribution systems. The source is not a limiting factor in any of the testing we are doing.

We've already mentioned 3D delivery systems that cut the vertical or horizontal resolution of the image in half before it ever gets to the display. If they are being used as source material you might not notice any loss between a full and half resolution display. The image quality isn't there at the input to the display.

The example we were asked to look at as a proof of performance of our ideas was LG's passive approach to 3D in their LW5700 series LCD sets versus Samsung's D8000 active 3D LCD display.

Since we suspect any set's 3D capability will be limited by its 2D capability we first wanted to look at the resolution capability of each set in 2D. The majority of image resolutions test patterns we applied are available from the Blu-ray disc of Digital Video Essentials HD Basics. Descriptions of those patterns are included at the end of this review. In our case we used uncompressed versions of the patterns from the Video Forge generator. There was no noticeable difference in the 2D performance, as seen on either of these displays, between the 2D signals from a good Blu-ray player and the generator.

Notes on what we might expect.

What you see in displayed resolution on a monitor is a combination of the circuits driving the display and the display itself. The test patterns JKP uses can often pinpoint where resolution is being lost, in the circuits or the display itself, if it is in fact being lost. This is important in determining if such a loss of resolution is easy to fix.

Our procedure for determining the real resolution of a monitor is to first set the Sharpness control at a point where there is no ringing introduced in the image. This helps us understand what the circuits are doing. We observe the condition of image enhancement as we adjust the Sharpness control. The Overscan pattern is displayed on the monitor as a reference for seeing image enhancement. It has both vertical and horizontal transitions to help evaluate what's going on in either direction. Some Sharpness controls, as is the case on the LG display, will soften the image as well as add ringing to the picture. Care is taken to get as sharp an image as possible without introducing any ringing. That setting on the LG set is 25-26 in the 0 to 100 scale of the control. The LG Sharpness control seems to alternate increases in horizontal and vertical enhancement as the control was turned up. At no point did we see both directions increase in sharpness with a single point transition in the control. The alternating change in enhancement can also be observed by displaying the SMPTE RP133 pattern.

Once at position 25 of the Sharpness control on the LG set we switched to the Pixel Phase test pattern to look at the video circuit frequency response. There was little amplitude variations visible in any of the rectangles. This is a clear indication of serious frequency response issues in the video circuits driving the display. Basically, at a position of no visible edge enhancement there is no visible resolution on screen from one-third system bandwidth on up to full bandwidth. Anything above the 25 position of the Sharpness control introduces edge enhancement.

We'd like to point out that using the Pixel Phase test pattern to determine frequency response is a challenging test, but it more clearly represents picture image requirements than test patterns that individually test horizontal and vertical resolution. It's a power bandwidth issue. In analog terms less power is required to convey a single frequency than say pink noise of the same amplitude. Circuits will often fail faster when they are asked to pass lots of information instead of a single piece of information. The Pixel Phase test pattern has horizontal and vertical information all at the same time, as does picture content. We believe it is a better test of the video circuit's capability than using patterns that separate horizontal and vertical information. We often see where sets can do far better in vertical and horizontal detail when each parameter is tested separately than when both are tested together.

Continuing to increase the Sharpness control on the LG set provides additional signal amplitude in horizontal and vertical detail to the 50% point of the control. Turning the Sharpness control higher pushes vertical amplitude back down while the horizontal amplitude continues to go up slightly. This suggests to us that we've reached the limit of the display itself in vertical resolution.

At the peak amplitude of vertical response, the 50% position of the Sharpness control, separate vertical resolution tests show single line transitions to be about 60% of full amplitude.

In short, the 2D capability of this display falls significantly short of being able to do full 1080p in the vertical direction. At the 25 position of the Sharpness control, where there isn't any visible edge enhancement, the set can't display much detail at one-third system resolution, let alone anything higher.

The Samsung set exhibits full resolution capability in both directions with no image enhancement with the Sharpness control set a zero.

2D Test Results

Sharpness control at 25 on the LG set, a position of no visible image enhancement. Note the lack of visibility of both horizontal and vertical information at full bandwidth.

Sharpness control on the LG set at 50. With ringing being introduced into the picture we still aren't at full amplitude for either horizontal or vertical resolution.

Sharpness control on the LG set at 50. Ringing (white edges around black lines) is being introduced into the picture.

Pixel Phase test pattern, Sharpness at 50% on the LG set. It's clear the display itself can show the resolution. The drive circuits have to be a significant limitation. Serious edge enhancement has to be introduced into the picture to get to this level.

Vertical Multiburst shown on the LG set. Here we're again showing vertical amplitude is clearly down at the peak amplitude capability obtained by setting the Sharpness control at 50%.

The Samsung set has no resolution or image enhancement issues in either 2D or 3D. The color seen in this picture is an issue of a beat between the TV and the camera.

3D Test Results

Knowing the LG display doesn't do well in 2D, we moved on to 3D. As the pictures show, displayed image resolution gets worse in 3D. As an example, diagonal lines became jagged in 3D instead of smooth in 2D. Even more vertical detail is missing. You need a sharp edge in the picture to see it. Looking at what we saw it is easy to understand how observers could say there is little or no difference in image quality between 2D and 3D on the LG set. They both fall significantly short of 1080p.

The LG set has a separate Sharpness control for their 3D mode. Its default position is much higher than the default for 2D. Edge ringing is far more visible. Taking that Sharpness control down to either the 50 or 25 positions shows less resolution in the 3D image than is visible in the 2D image.

SMPTE RP-133 in 3D on the LG set. Full 1080p vertical resolution is zero. There is nothing there.

Monoscope pattern on the LG set. This essentially shows that vertical resolution above 500 lines is gone.

Monoscope on the Samsung set in 3D mode. No loss in vertical resolution for a 1080p signal.

Conclusion

The LG set isn't capable of properly display a 1080p signal in 2D or 3D. Any claims made for little or no difference seen by viewers between full resolution 2D source material and half resolution 3D content is at least partially the fault of the display not being able to do a good job of either format.

The Test Signals

Overscan

Function: The primary function of this pattern is to check the amount of image area displayed on the screen. It also is used for checking rise time capability of a display as well as ringing in video processing.

Pattern Layout: There is a lot going on in this pattern. There are 1% markers on the horizontal axis and vertical axis that go into the picture by 25%. In addition there are markers on the diagonals at 2.5% intervals. Each direction is labeled with a letter so that overscan and centering information can be measured in every direction. The background of the pattern is a 50% gray with transitions either going towards white or towards black. The transitions have various rise times that are designed to aggravate video circuits into ringing, if they are sensitive to such rise times.

Individual pixels markers are proved at the center of the outside edges of each side of the pattern. The total pixel count is 1 to 10 from the outside edge of the pattern.

The Safe Action and Safe Title positions of 5 and 10% in from the edges are a holdover from standard definition. We by no means intend to imply that these numbers should be applied to high definition images. Our position is the 2.5% is generous for both Safe Action and Safe Title. In observing HD program production practices we find many people are treating the entire image area as being available, which translates to 0 overscan.

Descriptions of Use: The Overscan pattern is designed to show how much of the total 1920 by 1080 image is being displayed. It is also good at showing the presence or lack of edge enhancement. If present you'll likely see the edge enhancement in the single pixel level changes from 50% to 90% or 50% to 10%. Edge enhancement may be tuned for slower rise times so you'll also need to check the areas around the three pixel rise and fall times in some of the vertical back bars in the pattern. Information is provided in the following drawing on some of the transitions in levels. It's important to note that the sharp rise times are accomplished in a single pixel transition.

Pixel Phase

Function: Establish a one to one relationship between the input signal pixel count and the display pixel count. This pattern only works if the pixel configuration of the source exactly matches the display so this pattern is only useful on a 1920 by 1080 display. For this reason we are not including an image of the pattern in this text.

Pattern Layout: The background is about a 30% gray. (We say 'about' because there is no step in an 8 bit word video signal that hits exactly 30%. We are at bit 81, 29.68%. Bit 82 is above 30%.) The dark to light transitions run from a low of 5% to a high of 55%. There are six rectangular blocks of pixels, three on the left and three on the right. Each is out of vertical phase with the other. Starting from the top, the first two groups of information, top left and top right, are 2 pixel rise times. The left side is out of horizontal phase with the right side. There is a horizontal rise and fall time of about an eighteen pixel rise and fall time at the horizontal edges of the rectangles. It takes that long before the pixel transitions reach their ultimate amplitude. In this set of rectangles the pixel sequence is 5%, 30%, 55%, 30%, 5%.

The next rectangle down on the right side is one pixel on, one pixel off. The amplitudes are the same, 5% and 55%.

The amplitude excursion was kept to 50% of full video to keep the rise time within what we would expect of circuits without distorting the image. Such distortions would take away from the functionality of the pattern in testing for pixel phase.

The last three on the bottom are three phases of three pixel transitions. It takes three pixels to get from a starting point of 5% to a finish of 55%. The in-between steps are 18% and 41%.

It's important to note that this pattern test horizontal and vertical resolution at the same time. There are several display technologies that can do either one or the other well but not both at the same time.

When we first created this pattern in 2001 we had no idea there would eventually be displays that could easily handle single pixel transitions from 0 to 100%. By today's standards this pattern should not be a challenge, but it often remains a significant challenge so it has yet to be revised.

Descriptions of Use: As we've pointed out, this pattern is only useful as a pixel phase test pattern at its native rate. The 1080p pattern needs to be used on a 1080p display. Conversion to another rate will not create the one to one pixel relationship required for this pattern to assist in determining the pixel phase function.

In another application these patterns will provide a look at what is happening at the upper end of frequency response. The single pixel rise time wedges in the middle of the pattern represents a change in state from pixel to pixel or about 37 MHz in analog bandwidth. The single pixel transitions represent full bandwidth, the two pixel transitions, half bandwidth and the three pixel transitions, one third bandwidth. The single pixel transitions may appear to be slightly brighter than the others on a good display because the transitions to black are not as visible in the face of the adjacent pixel being bright.

The rectangles on the flat background have been know to cause streaking in the image, a fault of the display, not the test pattern. The high frequency, one pixel on one pixel off, rectangle on the right side of the image has also been known to cause image fold over in some video cards. You'll see an amplitude shift on the right side of the picture. It has been determined that this is a video output card driver issue in the cases we've seen. It is not the pattern itself, beyond it aggravating a set of circumstances that should not exist.

SMPTE RP 133 Resolution Chart

Function: The pattern describes image dimensions, resolution and black and white image limits. It was initially designed for medical diagnostic imaging tests but has been adapted in any number of high-resolution applications. An image of the pattern is not included here because of image resolution limitations of this document.

The original intent of this pattern is fully described in the SMPTE- RP133 document. In our case we are using it to illustrate image focus and detail as well as testing for chromatic aberrations in the lens of the projector. The gray background is helpful in assessing flat field uniformity. The black on white bar and the white on black bar are used to look for image streaking. The crosshatch will provide a quick look at image geometry. The letters and numbers will provide a reference for image focus at the outside parts of the picture.

We've separated horizontal and vertical resolution, both in full 100% amplitude transitions and shallow transitions. The steps are full, half and one-third resolution as represented by square waves so the rise times are always a single pixel. The shallow transitions are represented by one, two and three bit transitions in the 8 bit video word.