LCD motion blur: Eye-tracking now dominant cause of motion blur (not pixel persistence) - Page 2

I am referring to the middle graph you posted:



I'm referring to the concept of motion-compensated subfield dithering -- and essentially asking if FFD uses that at all -- e.g. say, creatively positioning and timing the dithering along the motion movement vector, so that you eliminate motion blur without needing to compress the pulses nearly as much. Like doing motion interpolation, but applied to subfields. Like short pulse at position X, then longer pulse at position X+1, even longer sustained pulse of output at position X+2, and so on. But that would complicate with the need to start the same gas cells with shorter pulses and ramp up to longer sustained illumination like in your graph, since you're now pulsing different gas cells when attempting to ramp up to the longer bursts of illumination -- limiting your ability to do this?

Picking your brain. You are the plasma expert in this thread, and I'm interested in some of the technicalities behind advanced plasma motion-blur reduction techniques, including the subtleties behind FFD.
Edited by Mark Rejhon - 11/13/12 at 10:46pm

Great, a confirmation from one of the plasma knowledgeable members here. Conceptually, it's similar to interpolating 600fps, and then basing the subfields off that, to keep the dithering along the motion vectors as much as possible.

Do you have any information about specifics about how motion compensation can interact with dithering - how much you can prevent the colorspace from degrading during fast motion, when using motion-compensation during sub fields? (slight increase in dithering versus massive increase in dithering). I would imagine that low contrast moving edges (e.g. bright gray edge chasing a slightly darker gray edge) would have no increased dithering because the cells are already 'hot' -- while a bright edge chasing a black edge, would have greatly increased dithering at the moving edge because the cells are cooled-off and need to be ramped up in illumination burst length yet (per xrox's diagram). Related question -- can plasma cells be ramped up from a cold state (pixels that have been off for many frames) to a very bright white 0.4ms long burst of light (FFD), without any pre-pulses needed, given the right plasma driver?

I know far more about LCD displays than plasmas, but can appreciate the science of plasmas.
Edited by Mark Rejhon - 11/14/12 at 5:43am

Interesting. I had been wondering about this. So, this would mean, that nowadays: Right, that's a good reference.
That said, nowadays (combined with Panasonic's drive system) that would mean, theoretically, you could use the motion interpolation techniques too, to increase temporal dithering while keeping motion blur down. (That said, it'd have the disadvantages of interpolation, such as increased input lag and potential interpolation artifacts, if this is done by any plasma.) Do you know if any plasmas do this nowadays?
_______

Funny how we're doing lots of plasma talk in a thread about LCD displays. So I will attempt to slowly steer this thread back onto topic.
Edited by Mark Rejhon - 11/15/12 at 4:50pm

Recently, I've pointed a point-and-shoot camera running at 1/600sec shutter on my LCD monitor (which was adjusted to maximum brightness and displaying continuous movement in a PixPerAn style pattern). It's not a high speed camera, but its live preview was taking fast samples of the sensor, apparently -- discovered this by accident on a Panasonic Lumix while holding the shutter halfway down while tracking the camera on a moving object onscreen. During the live preview mode on the camera (running at 60fps almost in sync with the computer monitor LCD), the object displayed on the preview LCD (showing an image of the object of the LCD screen that the camera was pointed at) -- the object was much sharper on the preview screen than the in the motion on the actual display the camera was pointed at! (This also happened to accidentally proves that my scanning backlight will have some major benefit even for an older "2ms" 60Hz TN LCD). The stroboscopic effect of capturing 1/600sec samples at 60fps (even if by accident, just for the Live Preview screen -- not being recorded), eliminated more than 80% of LCD motion blur, when I intentionally panned the camera manually (while in live preview mode, shutter half down) along the moving object onscreen. The same motion-sharpened effect will be observed under a strobed backlight. And this was just an old 60Hz TN display (Samsung 226BW computer monitor from 2006). Not a modern 120Hz display. I've become pretty confident I'll at least be able to eliminate 95%+ of motion blur, on a modern 120Hz-native computer monitor display, when running in full-strobe (non-scan) mode -- no backlight diffusion worries in the full-strobe mode.

Now, this brings some two (strange, experimental) ideas:
-- Modifying a shutter glasses transmitter to make LCD shutter glasses run in 2D operation: Briefly open and close both shutter quickly. This provides the short impulse effect, instead of using the backlight. Although this is not a practical method due to the dim image, it's an interesting experiment one could theoretically cheaply try without needing to do any modifications to a monitor. (It could be as simple as recording the IR sample, modify the IR sample, and play it back to the shutter glasses receiver in synchrony with a refresh, or it could be an Arduino electronic circuit modifying the shutter glasses signal in real time).
-- (more practical): Benchmarking motion blur with a camera tracking moving objects (like eye tracking). Build an Arduino-powered camera rail that is programmed to precisely move a point-and-shoot camera at exactly the same speed as a moving object on the screen. This would be a reliable electronic way to simulate eye-tracking, and actually scientifically measuring eye-tracking-based motion blur. I took a few test photographs while trying to manually pan the camera along a moving object on an LCD display, and was (with some difficulty, in about 1 out of 10 attempts) able to catch an accurate representation of what the human eye saw when eye-tracking a smoothly moving object onscreen.

Building an electronic camera rail that slid the camera along the motion vector (horizontal for simplicity, and it's the common "scene pan" case anyway), would allow more precise scientific measurement of eye-tracking motion blur, and easily scientifically measure "Honest Motion Ratios". The rail would be programmed with the pixel pitch of a display, and a test pattern would run the motion at an exact pixel rate. A video reviewer or benchmarker would set up the rail in front of the display where an "X" was displayed on the display. Upon the trigger of a screen flash driven by the test pattern (white on-off), a photodiode detector in the rail, would immediately start the rail running, and a standard camera (e.g. Canon with a firmware modification) would take a slow 1/10th second photograph while the camera is exactly moving along the rail following the moving on-screen object (to an accuracy of less than 1 pixel inaccuracy). For example, an object moving in a test pattern at 960 pixels per second (16 pixels per refresh), would have 16 pixels of motion blur at 60Hz LCD (well, potentially 18 pixels if you include the 2 milliseconds of pixel response time -- but as we now already know, eye-tracking is the dominant cause of motion blur). It would be 8 pixels of motion blur at 120Hz, then 4 pixels of motion blur at 240Hz, and theoretically just 1 pixel of motion blur on a Samsung "CMR 960" display -- even in that slow 1/10sec photograph taken while the camera was electronically moving to follow the motion. The 1/10sec photograph would ensure that ALL tracking-based motion blur is caught by the camera. (We are all familiar how a slow shutter can make photos blurry due to camera shake). The moving object would have a fine-pixel-pattern that allows you to count exactly how many pixels of motion blur there was, when the moving camera photographed the moving object. at a slow shutter speed.

This would finally allow the invention of "Honest Motion Ratio" -- a measured motion blur measurement, much like measuring contrast versus manufacturer claimed contrast.

For example, a Samsung CMR 960 display may end up having a measured "Honest Motion Ratio" equivalence of 560. (e.g. instead of being a 960 rating, it could actually be a measured 560 rating -- i.e. exactly the same same motion blur as a theoretical 'perfect' 560fps@560Hz display -- if you measure that a moving object has 560/60ths less motion blur than a 60Hz display, as measured from the number of pixels of motion blur in the photograph taken from the slow-shutter camera on the moving rail -- 560/60 equals 9 -- if the motion blur trail was 9 times shorter than the motion blur trail on 60Hz LCD)

This "Honest Motion Ratio" could be a more universal benchmark that is more directly comparable between displays, than the arbitrary "1080 lines of motion resolution" found in moving test patterns on Blu-Ray (which IMHO, is mostly meaningless because there's no single universal standard moving test pattern). The "Honest Motion Ratio" (actual scientifically measured motion blur) would be comparable across all displays, and maybe could become a universal motion blur measurement, for all future magazine reviewers, and blog reviewers to use, if we could find a way to somehow cheaply develop instructions for a precision camera moving rail that can be built for less than $200. (e.g. $100 Canon camera with a CHDK custom firmware, plus $100 of hobbyist motors and Arduino electronics). It would need to move fast enough at a sufficiently precise speed, so it'd need to be easily calibratable (e.g. with a certain kind of verification test pattern)

It's probably too complicated right now to be reliably simple for testers (yet), but it would be nice if this was the universal standard of actual scientific motion resolution measurement -- it's far more universally comparable, much like measured contrast ratios. Imagine if some manufacturer decided to step up to the plate and manufacture this for reviewers, and also made it available for rental.

Such "Honest Motion Ratio" actual measurements would cut through the "hype" and "false marketing", by measuring actual motion blur benefit of plasma FFD, scanning backlights, black frame insertion, and other "display voodoo" (Some that works wonderfully, and some that don't nearly as much) Much like how reviewers measure contrast ratios to debunk exaggerated contrast ratios. And putting a honest measured equivalence number behind those "CMR", "XR", "SPS", "2500Hz FFD", etc ratings.

The PixPerAn (motion test software) "chase" test is a similiar concept -- you narrow the gap until you've got the chasing moving object just barely touching the blur trail of the next moving object. This can also be used as a simpler method (but unfortunately more _subjective_ method) of calculating a "Honest Motion Ratio". The moving camera rail could certainly be set up in front of that test pattern or similar, for a more objective / scientific measurement of motion blur. The actual photographs being published online would be quite interesting to compare, even between different display technologies, CRT, LCD, plasma, etc.
Edited by Mark Rejhon - 11/15/12 at 9:41pm

xrox, I accidentally came across a Panasonic UT50, one of the 2500 FFD plasmas. It certainly looked very flickery for a plasma, so it was certainly doing very short impulses. So I can confirm a few things from visual observations:
(1) It uses motion interpolated subfields. I clearly recognized common motion interpolation artifacts. I imagine motion interpolation could be turned off.
(2) It definitely looked like it was squeezing impulses into a tighter time period than other plasmas; this plasma display was very flickery in a bright room. (It flickered like a CRT, moreso than other plasmas). In a dark room, it wouldn't be a problem - - flicker is good for reduced motion blur, anyway, if that's your goal.
(3) It had about a 5ms phosphor decay. There was a panasonic full-framerate moving ticker-tape demo that was moving text/graphics across the screen in 3 seconds (For a 1920-pixel-wide screen, that would probably be text moving at 600 pixels per second -- a conveniently programmed moving-ticker-text rate). I observed 3 pixels of trailing yellow-colored ghost (RG phosphor decay artifact). At 600 pixels per second, there's 10 pixels jump per uninterpolated 60Hz frame in 1/60sec (16.6ms). That translates to one pixel every 1.66 millisecond , so 3 pixels of yellow ghosting trail (red/green phosphor decay) is equal 5 milliseconds at that moving ticker tape text rate.

I'm actually quite surprised phosphor decay is still this bad, even today, in plasmas. The Panasonic FFD wasn't reducing motion blur as much as I expected-- the 5ms phosphor decay gives it only a honest, measured Motion Equivalence Ratio of only about 200 (1000ms / 5ms = 200) -- only about 3.5 time sharper motion than sample-and-hold 60fps@60Hz. That's not even an 80% motion blur reduction. No wonder CRT still is superior to plasma. Nontheless, this was the best plasma I've seen in terms of motion blur -- it had less motion blur than all the other plasmas in the store.

Back to topic, the very, very good news: I've recorded enough data to confirm LCD panels (3D optimized) can easily have less motion blur than plasma (even Panasonic 2500Hz FFD) when using a 150-watt-per-square-foot impulse-driven LED backlight .... That's expensive, but doable for about $300 of LED's in a 24" monitor. My design goal is 95% motion blur reduction (0.5ms backlight flashes every 8ms refresh at 120Hz, or using 1ms backlight flashes every 16ms at 60Hz. full strobe mode). The video game motion quality will be stunning. The metric I won't be able to meet, is have as good grayscale/black levels as this plasma... which looked pretty good even on the uncalibrated UT50.
Edited by Mark Rejhon - 11/22/12 at 7:17pm

This is an old thread, but here's some new information to contribute to this thread. Manufacturers are finally starting to do this; adding a strobe backlight to displays! The new LightBoost LCD monitors, just recently released, have been observed to eliminate motion blur, just as I have predicted.

I have obtained two of these monitors for testing, the VG278H and BENQ XL2411T.
High speed video of LCD pixel persistence being successfully bypassed by a LightBoost strobe backlight:



It clearly shows that the pixel persistence is kept in total darkness, and the backlight is strobed only when pixels are practically fully refreshed. I've also developed a guide on how to force-enable LightBoost for 2D and desktop use, not just 3D. There's also another third party guide.

Blur Busters Blog Guide:
HOWTO: Zero Motion Blur on LightBoost LCD Monitors

TechNGaming Guide
Eliminate Motion Blur While Gaming With NVIDIA LightBoost!

It is noteworthy that these are TN panels, so color quality is not all that good. There are multiple third party reports of "It's like a CRT!" from a motion blur perspective, good for FPS gaming such as Quake Live or Team Fortress 2.
So at least -- zero motion blur LCD's have finally arrived for video gaming!
Edited by Mark Rejhon - 2/22/13 at 3:33pm

I own both a BENQ XL2411T and ASUS VG278H, both of which supports LightBoost.
Some info I would like to add:

.....Yes, judder needs to be managed. You need fps=Hz. LightBoost is only supported at 100Hz through 120Hz. So you need to run fps=Hz, such as 100fps@100Hz, or 120fps@120Hz. No judder, no motion blur, no ghosting, no coronas, no plasma noise, no RTC overshoots (especially on newer 1ms monitors. This is because both pixel persistence and RTC overshoots are kept in total darkness, and strobe occurs when pixel transitions are finished). As a result, it looks like CRT motion. The primary picture quality problem is TN LCD color quality and black level, but these panels have the most "perfect CRT motion" effect of any panel ever made; even beating out plasma (due to 5ms red/green phosphor decay).

.....Yes, 60fps@120Hz looks worse. I see it with my own eyes too. Yes, it also looks even looks worse than CRT 60fps@60Hz. borf's explanation is fairly accurate. Your eyes are continuously moving when tracking a moving object, so frame repetitions give you a double exposure effect at different positions on your retinas. This causes a doubled-up edge effect during fast motion (much like 30fps@60Hz, but at halved distance between the edges) and is reported by some users in the LightBoost thread on HardForum and Overclock.net forums. Multiple confirmations.

.....You gain a competitive advantage during online gameplay. The lack of motion blur allows you to identify enemies faster while moving around, circle strafing, and anything that causes fast pans. Multiple confirmations, not just myself. See quotes below.

.....Amazingly, "1ms" actually makes a difference compared to "2ms" because of less crosstalk between refreshes. (Good for 3D, but also good for strobe backlight 2D). That means if you want LightBoost, the best monitors are the ASUS VG248QE and BENQ XL2411T since those are 1ms monitors. Other monitors that have the strobe backlight feature (often poorly advertised): ASUS VG278H, ASUS VG278HE, BENQ XL2420T, Samsung S27A950D, and Acer HN274H. The list will be kept up to date in my LightBoost HOWTO.

.....Display-induced motion blur below human detectable levels. Yes, there's still a slight amount of motion blur if you do extreme tests. For the most part, motion blur on CRT and LightBoost monitors is below human detectable levels in normal gaming. Most motion blur you see is human-brain induced, with the display no longer adding its own motion blur from its own technological limitations (e.g. sample and hold).

.....Several games run at 120fps just fine. I'm using just a single nVidia Geforce GTX 680. Games such as Portal 2, Battlefield 3, BioShock 2, Team Fortress 2, Quake Live, Counter Strike, etc, easily manage to stay at 120fps (once fps_max is raised). With a 680 SLI, you can also get Crysis 120fps, too. These are still great games, even if older.

.....LightBoost strobes is between 1.4ms to 2.4ms depending on config, according to my photodiode oscilloscope tests. PixPerAn motion tests also agree; motion becomes more than 80% clearer compared to a regular 120 Hz LCD, and more than 90% clearer compared to 60 Hz LCD. I can even count single pixels and jaggy edges in objects moving at 960 pixels/second. You can adjust the strobe duration via adjusting the LightBoost setting in the OSD, between 10% and 100% (but not at "OFF" setting). The picture gets darker if you adjust to shorter strobes, but that's good if you're gaming in a dark room anyway. There has been many positive confirmations of LightBoost doing an impressive job of replacing a CRT (if you don't care about color quality).

.....You can turn off LightBoost. These are excellent general-purpose 120 Hz computer monitors anyway. Some like LightBoost, some don't like it. A matter of personal preference.

.....It includes the pros and cons of CRT including flicker, however;
- It runs at 120 Hz refresh, so flicker is not a problem to most people. (not all, but most)
- It eliminates over 90% of motion blur compared to a 60Hz LCD (1.4ms versus 16.7ms)
- Today we have panels that have nearly no crosstalk between refreshes (a requirement for 3D).
- It's the first strobe backlight that a few desktop PC gaming CRT die-hards put away their CRT's
Some people have put away their Sony FW900 CRT's now -- because it looks just like a CRT in motion (even if not color).

So if you want the world's sharpest motion during PC video gaming on a FLAT PANEL, and you have at least a Geforce GTX 680, you're pretty much ready for a LightBoost display. (NOTE: You do not get the color quality and black level benefit of CRT. Just the "perfectly-sharp-looking motion" benefit).
Edited by Mark Rejhon - 2/5/13 at 11:28am

Correct, but we're talking about enthusiast PC gamers with high-end systems. Many have disliked LCD motion blur during fast action. You gain a competitive advantage during online gameplay, when you can identify enemies faster without motion blur while moving. Some of these desktop PC gamers have specifically said they aren't as concerned as much about colors, as the LCD motion blur problem.

Case in point -- User "Vega" from HardForum.com keeps a high-end Catleap 2B IPS 1440p LCD (120Hz overclock) monitor most things because it has excellent color quality, but apparently also likes to play the games on his ASUS VG248QE TN 1ms LightBoost monitor, because it has over 7x less motion blur (Catleap 2B actual measured 10.1ms versus VG248QE actual measured 1.4ms as measured from motion test software). Keeping high-definition pixel-sharpness even at fast panning speeds (1000 pixels/sec) is a sight to behold during fast movement in twitch FPS games for some of these enthusiasts. Completely blur-free pans that look as sharp as a stationary image.

For several of these gaming enthusiasts, clearest possible motion is the #1 numero uno image quality criteria.

So it depends on your goals. Games? PhotoShop? Movies? etc.
Edited by Mark Rejhon - 2/5/13 at 11:06am

Home theater displays or computer displays? Regardless, I will answer the question from the perspective of computer monitors.

It has been common that backlight scanning only gave small, incremental improvement in motion blur -- e.g. 20% improvement, 30% improvement. Often, these are not noticeable. It is only recently that strobe backlights have been developed that actually provide dramatic elimination of motion blur (90% improvement). This is because of a new turn of development...

Your A650 is not a shutter-glasses 3D monitor, correct? Having "3D" makes a huge improvement to scanning backlight abilities even for 2D. The reason is shutter-glasses 3D more strictly requires panels to refresh faster. This is because pixels must finish transiting before the next refresh, or 3D shutter glasses would not have been possible. The improved pixel transitions made possible by 3D panels, hugely benefits scanning backlights when doing non-3D operation. Backlights can be flashed on virtually completely-finished pixels rather than partially-finished pixels. So if you truly want the zero motion blur effect on an LCD, you really need a good shutter-glasses-compatible LCD panel with pixel transitions that don't leak fully into the next refresh (crosstalk between refreshes).
Even if you're not interested in 3D; because such panels work _much_ better with scanning backlights.

Newer panels do make a huge difference; you Today, some 120Hz computer monitors made in the last 12 months, show a dramatic improvement in motion blur when playing PC games with the strobe backlight enabled. These 23" and 27" models use a full-panel strobe backlight at 100Hz-120Hz, which actually was measured to reduce motion blur by nearly order of magnitude (relative to a standard 60 Hz LCD), and beyond an order of magnitude in certain cases. The models that most successfully bypass the sample-and-hold effect, with a true measured Motion Picture Response Time (M.P.R.T.) of around 2.0 millisecond (and less) are the following models.

Via following "LightBoost HOWTO" (to enable the strobe backlight mode in 2D) (requires nVidia graphics card)

- BENQ XL2411T (best; near zero crosstalk, lowest input lag)
- ASUS VG248QE (best; near zero crosstalk)
- BENQ XL2420T
- ASUS VG278H
- ASUS VG278HE (more crosstalk than VG278H)
- Acer HN274H

Via enabling "3D mode" via OSD, which eliminates motion blur in 2D too (because of strobe backlight):

- Samsung S23A950D (slight crosstalk, slight added input lag)
- Samsung S27A950D (slight crosstalk, slight added input lag)

None use interpolation, and as a result, 100% PC video game friendly. You do need a powerful Geforce GTX 680 or faster to get 100fps@100Hz or 120fps@120Hz, as strobe backlight operation is not supported below 100 Hz on these monitors.

Recently, the BENQ XL2411T and ASUS VG248QE was measured to have a 1.4 millisecond MPRT, true impulse-driven, putting it in CRT league for a common medium-persistence phosphor CRT computer monitor. Since the backlight flashes for about that long when the LightBoost setting via OSD is at the 10%-20% setting. The picture is a bit too dim though, so at a more realistic brightness, the MPRT is about 2 milliseconds. Most 60 Hz LCD's are measured to have an MPRT's about 10 times higher than that (even when LCD pixel response is 2ms or 5ms) due to the sample-and-hold effect 1/60sec = 16.7 milliseconds.

There is a motion test called "PixPerAn" from prad.de that allows people to verify motion blur and trailing/ghosting artifacts. With a LightBoost monitor, when enabling LightBoost, all trailing artifacts are virtually gone on the BENQ XL2411T and ASUS VG248QE, to nearly unnoticeable levels. It also means nearly no crosstalk during 3D operation. It requires a magnifying glass to see them:

This is a moving image (BENQ XL2411T and ASUS VG248QE) with the LightBoost strobe backlight enabled.
If you sit at a normal arm's length viewing distance, no visible ghosting, no trailing, no coronas, no RTC overshoots.

(Credit: OCBurner from HardForum.com)

Magnification, you can only barely see it:

(Credit: imgur from overclockers.co.uk)

So, as you can very clearly see, pixel persistence ceased to be the motion blur barrier.
It is pretty apparent in these images, even though these are stationary-camera rather than pursuit-camera images.

Even at speeds of 960 pixels per second, you can see the individual pixels -- the pixels are no longer motion-blurred (at LightBoost setting of 20%). You can even perfectly read the tiny "I NEED MORE SOCKS" text even when the car is zooming from left edge to right edge of the screen (1920 pixel wide) in just 2 seconds. No LCD's have been able accomplish such motion blur elimination until recently.
(Caveat: The image does become darker with the strobe backlight operation, and colors quality is not as good)

Blogger References:
TechNGaming.com: Eliminate Motion Blur While Gaming With NVIDIA LightBoost
3D Vision Blog: Take advantage of Lightboost for 2D gaming
pcgameshardware.de: Lightboost strobe hack
The Blur Busters Blog (my blog)

Forum References:
Thread on HardForum (hugely popular thread)
Thread on Overclock.net (hugely popular thread)
Thraed on overclockers.co.uk

Alas, these are not home theater displays.
But it does answer the question, "Are there LCD's that have CRT-sharp motion clarity without interpolation?" -- which is finally a YES if you're using a computer and want to play video games (Games like Half Life series, Crysis, Counter Strike, Bioshock 2, Team Fortress 2, Battlefield 3, etc). I'm specifically talking about motion blur, of course -- not color.
Edited by Mark Rejhon - 2/22/13 at 3:35pm