Red One Measured Resolution

[Steve White]

It's always fun to read such a detailed discussion of experimental methods. There are so many components to any camera system that it seems trying to quote a target resolution for a RED ONE is an unattainable goal.

In effect, the RED ONE is just a camera body. And in my opinion, the best RED can do is describe the efficiency of their filter, sensor, demosaic and compression algorithms, as Graeme has so admirably done. With enough sensor processing knowledge, it is quite possible to predict the system response of the RED ONE to a perfectly defined reference image. Above and beyond the sensor and processing algorithms, the resolution of the RED image depends on the reference image.

Any given configuration of the camera is a trade off between lens design, alignment, focus, subject matter and lighting. So if you want to know the resolution of your system, you've got to shoot your charts. For most people, I wouldn't be surprised if alignment and focus are the limiting factors. Even so, with current RGB capture and display technology, some perceptible spectral information is lost in the process (shoot some violet flowers to see what I mean).

 

[Andrew M.]

I agree with you but all these experiments leads to the better tools and more improvements by those that like to create tools more than using them and then having satisfaction that the best artists use them.

 

[Darren Orange]

Graeme, You may have already said this but does this mean Epic will produce 4K from its 5K sensor?

 

[Dj Joofa]

Even so, with current RGB capture and display technology, some perceptible spectral information is lost in the process (shoot some violet flowers to see what I mean).

And,

You get much more information when your filters aren't particularly monochromatic.

Spiff, there are a number of practical concerns here.

(1) You don't want the filters with a very wide spectral response as they might move the primaries towards the white point and hence your effective triangle formed by 3 primaries in the CIE xy space is a small one with poor coverage of the overall space.

(2) Since there is only so much area of the color space that 3 primaries will let you cover, then which parts of the space to give more preference? There is a rather remarkable theorem that states in order for any of the following two to be true:

(a) the attainment of maximum saturation in relation to the closest approach to the corresponding spectral color for a material having a specified dominant wavelength, and fixed luminance factor / visual efficiency,

or,

(b) the attainment of maximum attainable brightness corresponding to any desired chromaticity of a colored material under a given illuminant,

is secured if the material has only ones and zeros in its spectral response, with at most two transitions between them.

In actual practise it will be hard to have a material with such a response, that fact further constrains us with our coverage of the color space.

The implications of the above theorem were not lost upon NTSC and they came to the conclusion that the fidelity of color reproduction is not equally important in the color space. One aspect is that there is evidence of relative loss of discrimination between color perception at lower luminance. Therefore, a color TV need not reproduce all colors with equal fidelity.

The colors around locus of 560 nm - 700 nm have relatively good visual efficiency under various illuminants, say A, and C. And that naturally guides the selection of primaries. Notice that the green primary in NTSC in not placed really high enough.

So you select your primaries carefully to give optimal coverage of the color space in some sense and then use higher bit depth to partition that triangle formed by those primaries more finely and there you go. You have a good system.

 

[cinemano]

actually thats one thing im puzzled about.. being that RED is sharper or better than 35mm film, why would EPIC be needed? We lived happily with 35mm picture quality for almost a hundred years..

 

[Nova Invicta]

Matter of opinion

Matter of opinion

Having witnessed first hand numerous shoots shot on Red and combined with film and some comparisons between the two Red does not outperform film and lets not forget as respected as Graeme is he has a vested interest. Red is remarkable for its cost point but their has never been a panacea product and likely their never will be.
Film has its place, digital has its place why do the folks on this site think that they need to somehow be superior or deface artistic choice in favor of technicians or lab coats. The BBC are dictating an IT workflow from people who have never shot a thing in their lives if Red users want to be put in this catagory of theorising everything then its a sad world.

 

[Tom Lowe]

Graeme, You may have already said this but does this mean Epic will produce 4K from its 5K sensor?

I think the answer is yes... or very, very close to yes. For all intents and purposes, Epic should be "true 4k."

I'm curious to know how Epic will match up against Dalsa's "4K" in actual measured resolution. Isn't Dalsa also using Bayer and isn't their camera also subject to these same basic principles of actual CMOS resolution?

 

[Steve White]

Spiff, there are a number of practical concerns here.

Given my idyllic background in physics, I'm not concerned about things like "colour correction" or "white balance", as I view these as distortions one can readily apply to a given spectrum. What concerns me are the accuracy and precision of the captured and displayed spectra. Every given photon has a frequency and a polarization, both of which I'm interested in. Since I'm not aware of a way to measure both the frequency and polarization of a single photon, I'll leave polarization out of it.

Firstly, there is the issue of filtering on capture, and using a limited number of filters to capture the spectral information. There are several approaches to dividing up the spectrum - one is to use absorbing filters, another is to use dispersion (prisms) and translate colour frequency into position. There are practical limitations to both, and our eyes use the former.

So far as spectral capture on the filters is concerned, there is additional information in the form of "cross-talk" with broad band colour filters, which can be recovered to some extent via post-processing. You lose this information with monochromatic filters. The practical issue is that you have to have to commit to post-processing if you choose to use the broad spectrum filters. If your algorithm is too simplistic (i.e., if you assume monochromatic filters when they're actually broad-band), the output will be garbage. It is not clear to me that the right algorithm or sensor exist yet.

Display is a different can of worms. Right now we use 3 additive primaries, and this limits the display's ability to produce a given spectrum. That said, the current 3 primary system does an excellent job and most people aren't complaining. However, there is no theoretical reason we couldn't use many more intensity-tunable monochromatic sources, either by filtering a broad spectrum source or by generating the colours actively.

Looking at the way colour is represented digitally I'm consistently surprised by the subjectivity of it all. It doesn't have to be that way. Colour can be represented in absolute terms. But like you said, there are practical concerns, and always will be.

Every once in a while though, someone has to reject the status quo, go back to first principles, check the approximations, and push the envelope in a new direction.

 

[Darren Orange]

I think the answer is yes... or very, very close to yes. For all intents and purposes, Epic should be "true 4k."

I'm curious to know how Epic will match up against Dalsa's "4K" in actual measured resolution. Isn't Dalsa also using Bayer and isn't their camera also subject to these same basic principles of actual CMOS resolution?

Based upon what I read Dalsa is a single Bayer CCD. I am also seeing that Dalsa is limited to 4000 x 2048. While Red Records a full 4096 x 2304. So?

At the 80% value for both cameras you get...
Dalsa: 6553600 pixels
Red: 7549747 pixels

Now does this confirm that Red's images have more detail to work with?

Thoughts?

 

[GlennChan]

What concerns me are the accuracy and precision of the captured and displayed spectra. Every given photon has a frequency and a polarization, both of which I'm interested in.

- There is a ASC group looking at some of these issues (or at least similar ones)

Our advanced imaging subcommittee is closely collaborating with the Academy Science and Technology Council to reevaluate the accuracy of the long-standing CIE 1931 “standard observer” color-matching process for identifying colors based on an additive color mixture of red, green and blue primaries. There are known issues with the 1931 viewer-matching functions, but we are attempting to discover how serious they may be on a practical basis. The important two-part article “The Color-Space Conundrum,” which appeared in the January and April 2005 editions of American Cinematographer, was a project of our Technology Committee. Written by AC associate editor Douglas Bankston, it does a wonderfully thorough job of explaining color-space issues, along with workflow practices both current and future. (It can be accessed online at the ASC Web site: www.theasc.com.)

Personally I don't think we are better off trying to get better color accuracy.

One approach is to use a RGBE Bayer arrangement so that the spectral response comes close to the CIE CMFs. Sony does this in some of their stills cameras. But end users don't really notice the additional color accuracy. And most people don't view on color accurate monitors anyways.

2- Colors in our physical world change anyways because of varying illumination conditions, so I think there is a degree of color inaccuracy we can get away with / we tend not to pay attention to small color inaccuracies as they naturally occur.

-----------------------------------------------------------------------------------------------

Firstly, there is the issue of filtering on capture, and using a limited number of filters to capture the spectral information. There are several approaches to dividing up the spectrum - one is to use absorbing filters, another is to use dispersion (prisms) and translate colour frequency into position.

I think it would be quite difficult to design a camera that would get a lot of spectral information.
Sensitivity, optical issues, resolution would all be issues I would think.

Post and workflow would be quite difficult.

Display is a different can of worms. Right now we use 3 additive primaries, and this limits the display's ability to produce a given spectrum. That said, the current 3 primary system does an excellent job and most people aren't complaining. However, there is no theoretical reason we couldn't use many more intensity-tunable monochromatic sources, either by filtering a broad spectrum source or by generating the colours actively.

There are supposedly some laser and diffraction grating displays possible. As well as wide-gamut LCD, and what Genoa is doing.

I don't think it will make a huge effect on image quality though... most sets sold nowadays are far from color accurate.

And there are some color inaccuracies that happen through a professional workflow.
-People shoot on film (definitely not color accurate), or video cameras with video knee and color matrix cheating.
-In HD 709 -->SD 601 downconversion, it would be extremely rare to convert the primaries and transfer functions between 601 and 709.
-There is a disparity between the transfer function used by the camera (e.g. 709, 601) and the display (multiple figures quoted / no consensus, but it would be whatever a CRT does... a gamma of around 2.4).

Looking at the way colour is represented digitally I'm consistently surprised by the subjectivity of it all. It doesn't have to be that way. Colour can be represented in absolute terms.

Color perception would be difficult to represent in absolute terms due to weird stuff like surround effects. (Unless you measure + represent what the surround is too.)

 

[jbeale]

There are supposedly some laser and diffraction grating displays possible.

The OLPC XO-1 laptop does this with its display in color mode:

"Jepsen has described the removal of the filters that color the RGB subpixels as the critical design innovation in the new LCD. Instead of using subtractive color filters, the display uses a plastic diffraction grating and lenses on the rear of the LCD to illuminate the colored subpixels. This grating pattern is stamped using the same technology used to make DVDs. The grating splits the light from the white backlight into a spectrum. The red, green and blue components are diffracted into the correct positions to illuminate the corresponding R, G or B subpixels."

http://en.wikipedia.org/wiki/OLPC_XO-1

 

[GlennChan]

Thanks for that link... very interesting.

 

[GlennChan]

It's almost like the anti-bayer display. But the pattern is GBBR.

http://wiki.laptop.org/go/GTK_for_OLPC

2- Color-wise, a diffraction grating can be interesting if you want to go over 3 primaries. Using the diffraction grating you can split colors up into individual wavelengths... pretty much letting the display show every color possible (and also to avoid metamerism issues caused by displays varying in spectral output). A 3 primary display can't do that.

 

[Andrew M.]

US Patent (one) for new continuous color readout sensor

US Patent (one) for new continuous color readout sensor

Spiff with your idyllic background in physics, and the other honorable members of this forum, may I have a feedback on this idea, please.
Very interesting approach. Both Camera sensor photo site and LCD screen display having identical characteristics.

Sensor:
Each photo site recording not only intensity of the light but frequency of the light (color)
Display:
Each display element reproducing not only intensity but also frequency.

Looks like displays are very close to achieve this (see above posts) , will camera sensor do it soon?

This gives me a new idea.
How difficult is to produce the monochromatic array of camera sensors with the readout of the frequency of incoming light?
With all these GHz ability of the silicone based active devices, aren't we far above the blue frequencies of light? Don't we need just 3 time higher switching ability to reproduce frequencies? (audio)

In the laser technology they can modulate laser beam via ultra fast gate almost with the frequency speed of the laser. (GHz)

What if we have monochromatic CMOS just black and white sensor but two of them side by side or one as a dot inside the other as a ring.

For better accuracy It could be the same sensor, for half of the time is used to measure intensity of the light (no gate triggering) for the other half of the time (identical interval) we use it to measure the frequency of the light. (with the gate triggering)

One just normal CMOS sensor sensing the intensity of the light and the other identical next to it but with the fast switching gate (on off) switching the sensor on and off very fast (GHz speed) per each duration of the frame.
Charge of the first sensor will be proportional to the light.
Charge of the second sensor will depend from the intensity of the light and the frequency (color) of the light.
Since we know intensity of the light from the first sensor we subtract this value on the second sensor, thus what is left is clear response curve to the color.

By knowing the response curve of such sensor to the color you can get accurate continues slope reading of the color this way.

 

[jbeale]

the frequency of visible light is around 500 THz

the frequency of visible light is around 500 THz

How difficult is to produce the monochromatic array of camera sensors with the readout of the frequency of incoming light?

As far as I know that is not possible with current technology. Visible light is typically considered to be the wavelength range (lambda) from 700 nm to 400 nm (which we see as deep red to blue-violet). Because the speed of light c is (approximately) 3x10^8 m/sec that means the frequency of visible light is f=c/lambda
or (3E8 m/s) / (7E-7 m) = 4.28E14 Hz or 428 THz (red light) up to (3E8/4E-7) = 750 THz (violet light). THz = Terahertz (10^12 Hz).

So if you have a silicon device that can cycle at 50 GHz, that is still 4 orders of magnitude too slow to do frequency counting of visible light. I believe a device that works on that timescale would be molecular in size (like the dye molecules, which absorb a given visible frequency band). Our semiconductor device technology is not yet at that level.

 

[Andrew M.]

Yes, it is in standard frequencies domain.
I have to check how they do gating in the high speed laser multiplexers.
Something to do with edge slope transitions.
Flipping the state very fast and measuring light on the slope of the switch curve.
Flipping the state of the laser sensing diode very fast gives you different response from the blue light and different from the red, etc.
By triggering ultra fast transition you can detect the frequency of the light and especially if it is downconverted so frequency is much lower to deal with.
Used in telecommunication in the laser switching matrixes and NWDM (narrow wave division multiplexers)

Back to the source material, let me read a bit on it.
http://www.laserfocusworld.com/disp.../Computing-with-photons-at-the-speed-of-light
See figures at the very end of the article.

http://en.wikipedia.org/wiki/Optical_downconverter
http://www.wired.com/science/discoveries/magazine/16-02/st_timescales

 

[jbeale]

I used to work at a telecom company which made EDFAs (fiber amplifiers) for long-haul optical networks using WDM. The fastest modulation rates (a few years ago, anyway) were 40 Gbit/sec per channel, with for example 40 channels (wavelengths) per fiber. In that particular application, at least, there wasn't any individual active electronics switching faster than 40 GHz. The test lab's scope which could show the 40 Gbps waveform to check bit error rate was a pretty expensive device.

I suppose you could imagine doing direct frequency counting of a heterodyne signal, that is you construct super-stable optical reference frequencies and form a sum-difference product with some kind of optical diode, then count the microwave or mm-wave difference freq. In fact I vaguely recall something in Physical Review Letters some years ago about just such an experiment, using a nanoscale device formed by a STM tip as the nonlinear optical mixer. If that is what you're thinking about, I believe that design is not useful for a practical imaging device in the foreseeable future. The wikipedia reference is apparently talking about a similar idea, with a nonlinear crystal.

Sorry for the thread drift, I think all this is pretty far from Red or any practical camera.