Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

True Resolution of Mysterium difficult to determine.

E

edvan

New member
Joined
Apr 3, 2008
Messages
2
Reaction score
0
Points
0
I need to understand the true resolution and capabilities of the Mysterium Sensor inside the RED Camera.
Please help me.

24.4 mm in Horizontal Direction
29 microns per pixel
0.029 mm per pixel
34.48275862 pixels per mm
841.3793103 pixels in Horizontal Direction

13.7 mm in Horizontal Direction
29 microns per pixel
0.029 mm per pixel
34.48275862 pixels per mm
472.4137931 pixels in Horizontal Direction

This indicates true resolution of just 841 x 472 pixels.
************************************************

4520 pixels
29 microns per pixel
0.029 mm per pixel
34.48275862 pixels per mm
131.08 mm in Horizontal Direction
25.4 mm / inch
5.160629921 inch in Horizontal Direction

2540 pixels
29 microns per pixel
0.029 mm per pixel
34.48275862 pixels per mm
73.66 mm in Horizontal Direction
25.4 mm / inch
2.9 inch in Horizontal Direction

This indicates true sensor size of 5 x 3 inches.
This is very unrealistic.

*****************
What am I to believe ?
 
Are you talking pixel linear dimension or pixel area (square microns) ? The Red One sensor pixel size is 29 square microns in area, that is a square 5.398 microns on each side assuming 4520 x 2540 pixels filling an active sensor area of 24.4 x 13.7 mm.

The pixel area of 29 square microns is just based on the pixel count and overall sensor size. As far as the exact pixel size, I don't think any sensor has a 100% fill factor, although with microlenses you can get pretty close.
 
Where'd you get the "29 microns per pixel" idea? Red's 5 microns per pixel.
 
29 microns square is the area of the pixel, not it's linear dimension, which is just over 5 microns. The sensor is 4900x2580. To clarify, that's the full Mysterium sensor, but not the addressable or recordable pixel area.

Graeme
 
Thank you for finding my error. Here are my new calculations

Thank you for finding my error. Here are my new calculations

Thank you for finding my error.
Here are my new calculations

29 sq microns
5.385164807 microns per pixel


24.4 mm in Horizontal Direction
5.39 microns per pixel
0.01 mm per pixel
185.70 pixels per mm
4530.97 pixels in Horizontal Direction

13.70 mm in Horizontal Direction
5.39 microns per pixel
0.01 mm per pixel
185.70 pixels per mm
2544.03 pixels in Horizontal Direction

4520 pixels
5.39 microns per pixel
0.01 mm per pixel
185.70 pixels per mm
24.34 mm in Horizontal Direction
25.40 mm / inch
0.96 inch in Horizontal Direction

2540 pixels
5.39 microns per pixel
0.01 mm per pixel
185.70 pixels per mm
13.68 mm in Horizontal Direction
25.40 mm / inch
0.54 inch in Horizontal Direction
 
Well the only what looks right about here is 185lp/mm/2 (you mixed numbers in vertical direction)
So Mysterium delivers 4096 unique values per horizontal line and needs lenses that can deliver 200lp/mm if you are pixel picky.
Using traditional TV broadcast technique of measuring things, determined experimentally, I have heard that RED camera (not just Mysterium) throughout the whole path out of the camera to the RAW file and then to the TIFF delivers throughout the whole chain of events 1800 vertical lines TV equivalent.

Well, I rather stay with horizontal pixels since vertical is different for different aspect ratio and many people from the RAW world knows how to judge the picture based on so and so many pixels of the Bayer CMOS sensor rather than TV lines.
 
Andrew, why are you jumping from pixels per mm to line pairs per mm? You're now out by a factor of 2! 185p/mm implies 92.5lp/mm which is well within the reach of good glass (And RED glass, from which I've seen it produce aliasing out at 4k when we've shot without OLPF).

1800l/ph is 3.2k, or about a 78% linear conversion factor from the 4k samples, which is significantly greater than the > 70% factor I quoted for a Bayer sensor when people were asking this stuff back on DVXUser. And We've read independent confirmation of this number on CML too.

Graeme
 
Yes, one has to specify exactly what all this numbers are coming from.

Confusion is in the fact that gentleman above was calculating maximum possible recording resolution of Mysterium and not 4K usual resolution that most of us use at this moment.

He probably took numbers from the RED website here:

http://www.red.com/cameras/technology

"Record 2540 progressive at up to 30 fps REDCODE RAW. With 4520 X 2540 pixels, Mysterium™ puts pure digital Ultra-High Def in the palm of your hand."


So 2540 X .7086 = 1800

So I am right on your numbers, you were thinking 4096 X .7 I guess.
Also I have seen 1800 number mentioned from few other sources, again they probably count all possible recording pixels, since right after 1800 number I have seen 70% there.
As to the Line Pairs Per Millimetre I put (divided by 2) there:
"Well the only what looks right about here is 185lp/mm/2"

1mm / .0054mm (pixel size) = 185.18 pixels per mm on Mysterium.

Confusion here is that I put/2 and 92lp/mm, it mean one black and one white line on the projector test that I made with MP lenses.
When we are talking lenses lp/mm, there is CA and sharpness factor.

Sharpness is not as crucial as CA. 50lp/mm equivalent sharpness will not make Bayer demosicing as much less accurate as 80lp/mm equivalent CA
OLPF is limiting resolving power of Mysterium to probably 50lp/mm but it acts on all colors at the same time in terms of exact location of the color sample so it takes care of very important factor of colors coplacement. 100lp/mm CA equivalent may offset color location by one pixel or so thus producing very bad color artefacts that will show in the TIFF file as a 2 pixel wide color fringing.
 
Andrew, you're post is very confused / confusing. It's hard to answer.

The sensor max rez we can see is 4520x2540. The max record area is 4096x2304.

If you record a test chart while shooting 4096x2304, through Redcode RAW, you can, upon analysis of the chart, see resolution out to around 3.2k, which is about 1800 or so lines per picture height. This is measured resolution, not theory resolution. I've measured this, as have others (see CML).

I don't know where you get your numbers from for the OLPF, but it's wrong.

Graeme
 
This quickly gets into what is referred to as The Tyranny of Numbers, or what my college statistics liked to refer to as the difference between Accuracy and Precision. One can sit with a calculator and attempt to define resolution capabilities based on theoretical limits and understandings, or one can actually shoot a test and measure the information directly.
 
To make it simple, as you always state, the max vertical resolution of the whole RED camera, not just Mysterium sensor, is 70% of the vertical pixel count used for recording.

Very simple.

As to OLPF I just made calculated guess.
Mysterium packs 185 pixels per millimetre.
I know you explained it in one of your posts using maximum frequency and minimum frequency cut of scenario.
But by simple not scientific logic, to avoid moiré we need to use more than one and less than 2 pixels for each sample.

So 50lp/mm it means 100 samples per millimetre as a bottom frequency of OLPF.
So we are packing 1.85 pixels per one sample for averaging. (it is 54% ratio from one)

Well you mentioned 70% (I don’t know if it is after sharpening or before) so I am wrong if it is before sharpening.
In such a case you are right, should be 65lp/mm as a bottom frequency cut of equivalent for OLPF. (185 pixels/mm X .7 = 129.5lines = 64.75lp/mm)

I prefer to imagine OLPF filter as a glass that makes projected image on the sensor bit out of focus rather than frequency limiting device, spreading sharp edges across 1.5 to 2 pixels. Somehow it sits better with my imagination.
Anyways, it is not scientific explanation, it is more logical one, my favourite.

OLPF link
http://www.epson.co.jp/e/newsroom/2007/news_20070906.htm
 
The 70% is the approximate amount of linear resolution retained through both the OLPF and demosaic process. I'm now measuring around 78%.

The OLPF acts as a filter to remove higher frequencies. You can think of it as a very small gaussian blur on the light. If you hold an OLPF up to your eye, I don't even think you can see the blur. The OLPF is always an imperfect filter in the sense that it cannot ever fully attenuate any frequency to zero, but it is set so that it has maximum attenuation before the 4k frequency point.

Graeme
 
Now I got it!

The final linear resolution is not only factor of image frequency after OLPF but also the debayer algorithm. I guess good debayer algorithm can reverse engineer what OLPF lost in terms of details through sharpening process.

Well I guess this 78% number will be getting higher and higher in time by improving debayer. Especially if you start using two frames simultaneously for debayer calculations.
 
Certainly you can add electronic sharpening to the image, but we don't do that in the debayer. We just extract as much real detail as we can. You can add some "OLPF Compensation" if you want, which is sharpening though.

Demosaicing is one of those things that you can improve almost indefinitely with more processing power / time / algorithms. But who wants to wait 10 minutes for a frame?

Graeme
 
My math must be off as well... I had E*E = > than E, that's assuming that E= enough. ;oP

I'm just being silly, I also privately obsess over lp/mm on film and horizontal resolution on video cameras. Given the current content delivery mediums & technology I really think we've crossed the threshold of sensor resolution doesn't matter as much anymore. Sure, if you're displaying on IMAX or doing some very specific VFX operations it may matter but those are specialty applications. Personally, when it comes to the sensor, I'd rather see advances in DR S/N ratio if anything (I have no complaints in that area though it's probably only going to get better).

From my experience, the MTF/contrast characteristics of the optical system (lens) matters more than resolution when it comes to percieved sharpness.

Many DPs will agree that they like to use diffusion when shooting 35mm to soften it up a bit. I have the same preference when shooting 35mm & up. Once we get through all our testing, I have a feeling that I'll go in the same direction with the Mysterium.

Graeme, first off your work is outstanding. I know you're busting your ass perfecting the softwaew but I want to mention something, If it's not already in the works, and I'm betting it's at least been discussed, I'd love to see hardware (PCI-e or external) assistance for dealing with Redcode. I can't help but wonder if the nVidia Tesla or QuadroPlex, and the other systems out there are potential candidates to offer significant speedups. Even a good 'ol renderfarm setup sounds appealing to me.

Perhaps when you finally get that bathrom break you've been loking forward to for the past couple years...
 
The advantage of higher resolution really comes in because we're using a sampled system, and that involves filtering to avoid nasties. It means we can filter properly in the optical domain, and if needed also do better filtering in the digital domain. Without the "extra" resolution RED offers, we could not do that. Resolution does matter, and it matters for both the visual effect of that detail, and the technical issues relating to filtering.

Graeme
 
A micron (µ) is 1/1000th of a milimeter (or one millionth of a meter), expressed in decimal: 1µm = .001mm

5.39µm pixels are not that small, but since the Mysterium sensor is big, it translates into high resolution.

The best lenses will resolve 200 lp/mm on the test projector = 2.5µm pixels, but the MTF response from the lens at that high a frequency is likely to be less than 10%. The human eye can detect individual black and white lines at about 5% MTF.

The key thing here is: A lens that has visible response at 200 lp/mm will have good usable response at 68 lp/mm.

92.5 lp/mm [max, no OLPF] * 0.78 = 72 lp/mm effective resolution.

This is PLENTY. ARRI tested the latest filmstocks by scanning them at 4K on the ARRIScan, and color negative film tops out at 80 lp/mm.

I didn't even think this was possible at all several years ago. And when I did think it was possible I still thought it would be many years away.

Graeme, thanks for the info, hopefully I will see you again at NAB this year.
 
You must log in or register to reply here.
Back
Top