Video Calibration

The New Wave of Color Calibration Technology: Cube Calibration

ARTICLE INDEX

Introduction to Cube Calibration

One of the Awards given out in 2012 by Secrets of Home Theater and High Fidelity for "Technology on the Rise" was for Cube Calibration. For most people this probably doesn't mean much yet, but for those of us that are calibrators, or enthusiasts with an interest in calibration technology, this was a huge development. One might even say that Cube Calibration is the biggest improvement to consumer display technology since the introduction of the Color Management System (CMS).

A traditional CMS calibrates between 5 and 16 points. Usually 2 or 10 grayscale points, and 3 or 6 color points. When designed and implemented correctly, these controls can lead to a very accurate image, with a very color-neutral grayscale and very accurate color primary points. We publish a graphical representation of this, which entails a stck target representing various correct color points; this is known as a CIExy 1931 color diagram, and accompanies all of the display reviews we and many others do. You may also assume that when those measured points match with the reference points on the chart "This display has perfect color", but this is actually a common misconception, and one that Cube Calibration helps to address.

As we review cube calibration technology from Lumagen and SpectraCal, first we have to review how current calibration technology works, and the issues that exist in it.

SPECIFICATIONS

SpectraCal ColorBox

  • Video Inputs: 1 HDMI
  • Video Outputs: 1 HDMI
  • 3D LUT Memory Settings: 6
  • MSRP: $1,595 USD

Lumagen Radiance Mini3D

  • Video Inputs: 2 HDMI
  • Video Outputs: 1 HDMI
  • 3D LUT Memory Settings: 8
  • MSRP: $1,499 USD
  • SpectraCal
  • SECRETS Tags: Cube, Color Calibration, Video, SpectraCal, Lumagen

 


Current CMS Systems for Cube Calibration

First, let's look at the unmentioned issues that exist from the CIExy chart that everyone sees attached to reviews. The first issue, and a large one, is that color is actually a 3D object. In the CIExy chart that most people use, colors are represented by two, not three, distinct variables: x and y. In this system, x and y values represent tint and saturation, while Y represents brightness or luminance. As you can see we have accounted for hue and saturation but not the Y value, which happens to be the most important element for our visual perception of colors. Even if all of these points line up perfectly, given there is no representation of luminance on these charts, highly perceptible errors can still be observed. For example: if there is too much red, green and blue, then red will be too bright. The shade of red might be perfect, but the Luminance is just too high. This is one fault that the common CIExy chart does not show.

The other major issue with the CIE diagram for evaluating color is that you are only looking at 3 or 6 points in total. With our standard 8-bit RGB displays, we can generate 16.7 million colors, but we're only looking at the performance of 6 of those to determine the performance of a display. What about the rest of the points? The CIE diagram represents a snap shot of a small sliver of data as the data is all collected at a given stimulus point. What if 100% saturated red or blue are perfect, but every other saturation below 100% is incorrect? What about colors that involve mixing red, blue, or green together, and not just a single primary color? The chart shown with most reviews addresses none of these questions.

We can show data that details this performance better using a few different items. One is a saturations chart, which measures saturations other than 100% to see how well the CMS handles those colors. Another is a luminance chart that will show how less intense signals are managed by the CMS. Finally we can use the Gretag Macbeth Color Checker chart that is well known in the photography world, but not as well utilized in the consumer display world. This chart uses 24 colors that represent common objects in the world, like natural greens, sky blue, skin tones, and more, and none of these are points that we try to calibrate to with a CMS. This is a much more demanding test of displays than other test patterns, but also more applicable to real life.

If you're only concerned with how well Cube Calibration performs, feel free to move onto the next page, as these next few paragraphs go into some technical depth about why current calibration systems don't often work correctly.

A key reason to understanding why CMS systems can't do this correctly is to understand how most CMS systems work. Your display outputs everything as RGB values. LCD, CRT, Plasma, OLED, or any other display technology out there currently uses RGB for its display output (although a few add a yellow "primary" to increase light output, but this can cause problems obtaining correct color performance). Most CMS systems are instead designed around HSL, or Hue, Saturation, and Lightness controls, and not RGB controls. When HSL was designed in the 1970's, it was easier for computers of the time to process and for people to comprehend, and for people visual selecting a color; it makes it easier to do than to try to calculate RGB coordinates. Making that orange lighter, or darker, or more saturated is a single control and much more convenient that using RGB.

Where the RGB color space is a linear cube, the HSL color space resembled a pair of cones, with one inverted and the two cones connected at their bases. To go from Black to a solid color in RGB is a linear line from one corner of the cube to another, whereas in the HSL colorspace the path is not as linear. The result of this is that while those points on the CIExy chart might be correct, the points in-between are much harder to determine when calculating them in HSL than in RGB as the colorspace isn't linear like RGB. In many traditional CMS systems this becomes very apparent when you test for saturations, as you can get 100% or 75% correct, but any of the other values are off by a considerable amount, as it can't correctly calculate the non-linear HSL colorspace correctly.

Both the Lumagen Radiance series and the SpectraCal ColorBox use a linear-Gamma RGB color management system for their internal calculations. This should allow for them to produce more accurate colors than the HSL method when you look at data inside the primary and secondary color points. Of course, they also can support 3D cube calibrations, which is what we will discuss next.

3D Cube Calibrations

As we mentioned, most CMS systems are based off the idea of calibrating the white point, then calibrating the primary, and possibly secondary, colors, and then using calculations to get the other millions of colors correct. This is very nice in theory, but it has some problems in real world practice. We might have a display that doesn't exhibit a perfectly linear response, so even if the calculations are correct, the resulting output is incorrect. The calibration might require large adjustments in the CMS system, and the bit-depth available for calculations is too small, resulting in posterization or banding. The controls might not have enough adjustment available to get everything correct, or they can interact with each other, resulting in a tricky balancing act for performance.

Cube calibration works differently. The way to visualize this is to see color as a 3D cube, with 8 points, one at each corner. Two points on diagonally opposite corners are White and Black. The other 6 corners are the primary and secondary colors. Inside the cube you can encompass all of the colors that are possible to be created using RGB. Our previously discussed color systems use between 5 and 26 points for calibration. 6 points are the corners of the cube, and the other points are a path from black to white inside the cube, representing the grayscale. To determine any other color, you need to calculate the distances between these known points, and then output that value.

What a cube calibration does is take multiple measurements for each side of the cube and inside the cube, not just a single corner measurement. With the Radiance line, we can do 5 points on each side of the cube and between each corner, for 125 total points of data, plus 21 independent points along the gray-vector. The ColorBox allows for up to 17 points per side, or over 262,000 data points. Now instead of having potentially large distances between known values, we have very short distances to transverse.

This lets us now account for non-linear displays, and it substantially reduces the amount of possible error in our calculations. Now using more complex methods of checking display performance, such as the Colorchecker chart or Saturations chart, we should start to see lower values in those data points. The color points that most people use for reviews, primary and secondaries, will not have changed if the existing CMS was already good enough to make them accurate. What you will see from this review is that just using those points doesn't provide a full picture of what a display can do, and just were CMS systems are lacking.

 


Cube Calibration In Use

Test Setups and Calibration

Testing was done with two totally different setups to fully evaluate how the different products perform. Chris Heinonen tested the SpectraCal ColorBox using both a Sony VPL-HW50ES Projector which has a built-in CMS, as well as a Sony KDL-32EX308 LCD TV that contains only a 2-point white balance control. All of his calibrations were done using CalMAN 5.0.4 or 5.1, with an i1Pro and C6 meters.

Mark Vignola used a Lumagen Radiance 3D XS video processor with a JVC RS25 projector and a Panasonic VT30 plasma display. Calibrations were performed using CalMan 5.1 software, along with an i1Pro2 meter. Cube calibration is available in CalMAN as well as the Chromapure calibration software packages.  Chromapure however, only supports the Lumagen Radiance and not the ColorBox.  In order to reduce the number of variables, we chose to both use CalMAN.  We will report back on Chromapure's interface in a later review.

Beginning with the Sony VPL-HW50ES, the SpectraCal ColorBox allows for up to 262,000 points of data, but that is certainly overkill for our use. Instead I tested it with 729 points and 4096 points, to see if there was a large difference. For both of these settings I first got them as accurate as possible inside of the Sony CMS without making large adjustments. Large adjustments lead to irregular display responses, as we discussed earlier. Some would actually suggest no adjustments beyond white point, but the Sony already had very good response at different saturations so I just left it calibrated.

As the ColorBox does nothing besides color correction, all you need to do to use it is place it in your HDMI chain before the display, and then select one of the six memories using the remote. Then in CalMAN you choose which memory you are going to use, and use their 3D LUT workflow. I chose a gamma of 2.2 and the standard Rec. 709 HDTV colorspace. I also chose a target dE of only 0.5, which is much lower than the default. This will cause it to take longer, but should provide more accurate results. Calibration of the grayscale for each took around 15-20 minutes using the C6 colorimeter. Calibration of the color was a much longer process.

The 729-point cube took around 3.5 hours to complete. The 4096-point calibration took almost 8 hours and over 10,000 readings to finish. Happily I started this before going to bed one night, so it was easy to let it complete in the dark without being disturbed. The C6 is a fairly fast meter with CalMAN, but for doing a huge number of readings like this I would really like an even faster one. I certainly wouldn't use a spectrometer as it would be very slow. Once both were completed, with different memory locations, I was free to switch between them to compare.

Calibrating the Sony HDTV is much more limited in setup before calibration. All I could do for it was pick the most accurate color temperature and then adjust the white balance. Everything else was left at the defaults, as there is no CMS to use to adjust it. I only used 125-points for the Sony, mostly wanting to see if the ColorBox can take a display that costs 20% of what it does, and turn it into something with far more accurate color. Thankfully running the calibration on the Sony only took an hour with that few points.

In Mark's set-up, the Lumagen sits as the last piece of equipment before the both the plasma and projector. Both the Panasonic VT30 and JVC RS-25 have fairly full complements of calibration features, including CMS and grayscale controls, but Lumagen suggests its users only select a color pre-set that gives oversaturated colors, and then do nothing more than set brightness, contrast and high grayscale. Once these were completed, Mark conducted the CalMAN 3D LUT calibration workflow exactly as Chris did. CalMAN also allows users to select which CMS we'd like to calibrate: the Lumagen possesses 4 memory positions, but can accommodate 8 different full CMS calibrations. These can be mapped, as the user sees fit, to any number of different situations. Unlike the ColorBox, the Lumagen is limited to only 125 points of color correction, plus 21-points on the gray-vector, so no selection is made by the user regarding how many points are corrected. While Mark conducted calibrations on both the RS-25 and VT30, we'll only be discussing the results from the plasma for simplicity purposes.

The i1Pro2 is not as fast of a meter as Chris's C6, so automated calibrations took a little longer in his set-up, with Calibration of the grayscale taking about a half an hour, and 125 point color calibrations taking approximately an hour to complete. Either way, the process is fairly streamlined and painless once things get started.

Results and Analysis

Before the cube calibration, the Sony VPL-HW50ES was a fantastic performer. You'd have to be incredibly picky to want better performance that it could do out of the box with a simple calibration. If you are that picky and demanding, you can get even better with a 3D LUT added on. The gamma, grayscale, and CCT numbers all improved after the cube calibration, but not by gigantic amounts. That little bump of an error at 5% on the gamma is now totally fixed, though. When we move to the more challenging saturations and color checker tests, the average dE on the 50ES drops from very low, around 1.4, to obscenely low at below 0.75 dE 2000.

Practically speaking, if you look at the Sony VPL-50ES with a cube calibration, you are going to see a perfect, reference image without any real flaws. Colors, grayscale, gamma, and everything else is so close to perfect, you visually can't see the difference. The idea of seeing an image that is practically as good as the director of your favorite film is a bit crazy, but that is the crazy world we might be living in now.

Watching all of my favorite content, it was amazing how clear and detailed everything was. Skin tones were totally neutral, oceans and lakes looked more natural, and shadow details were much better with the improved gamma and grayscale. It was clearly the best picture that I've seen in my home, and possibly the best-projected image I've seen outside of a movie theater. I looked for all the flaws that calibrations can introduce, like banding and posterization and other bit-depth issues, and saw none of them. All I saw was exactly what was on every single Blu-ray I watched and nothing more.

I also switched between 9 and 17-point calibrations on the 50ES and saw no difference. I could move between them instantly using the ColorBox, but not once did I see anything that made a difference in my eyes. Because of this, I'm only reporting on the 9-point calibration, as the 17-point numbers were all almost identical, and really within the margin of error of the instruments. They were certainly within the <1 dE2000 level where you could see a difference side-by-side, and if I had both images up next to each other, I'm certain I would see the same thing on each. Perhaps if you work in a mastering studio you would need this, but in the real world, the extra 5 hours spent on the calibration made no visible change to my eyes.

With the cheap-o Sony TV, we did see a bit of a shift for the worse in the overall CCT, which is unfortunate. Aside from that, everything else improved across the board. Gamma went from way too dark to spot-on perfect, our grayscale dE2000 went from visible errors to absolutely nothing, and the color checker dE2000 error was cut in half. There was a bit of a rise in some of the saturations dE2000 numbers, but overall most numbers really improved, even on a display this cheap with no CMS controls. It didn't make it reference quality like the projector, but it made it much nicer than it was before.

In Mark's case, instead of testing how close we could get the VT30 before using its internal controls, we thought it would be interesting to test THX mode, which, according to THX: "offers the closest thing to a pre-calibrated setting out of the box. This reproduces the Rec.709 color gamut, luminance levels and other settings used by filmmakers in the mastering studio—with the push of a button." As you can see from the data we've presented, THX mode was decent, but not terrific. While some things aren't bad, nothing is perfect.

After the full calibration with the Lumagen, we can see a marketed improvement in all aspects of the displays performance. Grayscale now tracks very well, with an average dE2000 of 1.98, a significant improvement over what the THX mode was giving us (dE2000 5.6). While the gamma is substantially improved, these results could easily be improved by a few minutes of tweaking. Results on the saturations and color checker test are now an average of dE2000 of 1.6 vs. 4.9 in THX mode; a substantial improvement. Viewing results were equally as impressive as what Chris experienced. Blu-ray content looked superb: this summer's Avengers, a reference quality disc, looked stunning. The best way to say it is: everything just looked right.

We did note that the state of the display before calibration did seem to have an effect on the quality of results that we achieved. Mark started his calibration of the VT30 with absolutely nothing done inside the display except contrast, brightness and 2-point grayscale. All other controls were left at factory default – and while we did not show the data collected prior to calibration with the Lumagen, this starting point was actually worse than what THX mode provided. Alternatively, Chris's calibration of the VPL-VW50 started with an already calibrated display that measured very well even before external calibration was implemented. His results from the cube calibration were better than we saw in the VT30. What Mark achieved with the VT30 post calibration were more similar to the Sony LCD, which, as a result of simply having less calibration controls available, also started as a basically native, highly incorrect display.

While the point that you start at seems to affect post calibration results, this is likely a result of the cube calibration implementation in CalMAN software at this point versus the magnitude of corrections achievable by either of the external solutions we evaluated as all of the errors that we observed are well within the capabilities of being corrected by both the Lumagen and Colorbox. As the automated cube calibration matures, we anticipate that the starting point will become less important.

 


Conclusions about Cube Calibration

Displays have gotten much better over the past decade. CMS systems have gone from hidden away in service menus that you can't access, to having full controls that you can often adjust yourself. At the same time, these controls are not always intuitive and you can often have a situation where you can adjust them to give yourself "perfect" results in calibration software, but have awful results in reality. We have also been given more tools now to evaluate and examine displays in depth, allowing us to find issues we couldn't easily report on before.

Thankfully Lumagen and SpectraCal have risen up to the challenge with their 3D LUT devices. In testing we found that they are able to take an image that is bad and make it good, but more importantly take an image that is great, and make it so close to perfect that you can't tell the difference. With these tools you really can see images at home that are more accurate than ever before, and see exactly what your films look like. Not close, not within a certain acceptable range, but really perfect.

Choosing between the two different setups is a bit harder, but for most people the Radiance will be the one to choose. The Lumagen offers additional video processing features, multiple inputs, and more than the ColorBox and it's available in its cheapest version for less than the ColorBox. The ColorBox can do far more points, and so if you are incredibly exacting you might demand that. With projectors I believe this is overkill, as the lamp will age fast enough that the extra 6-7 hours spent on the calibration each time will be a larger pain, but with a flat panel it is more reasonable to do.

Whatever route you take, having a Cube Calibration and 3D LUT in your system will give you a far better picture than you currently have, and can be added to any system out there. You might just forget 4K and its additional resolution potential you might not see when you can improve the equipment you already have and achieve a different you will notice.