Home · Register · Search · View Winners · Software · Hosting · Software · Join Upload & Sell

Moderated by: Fred Miranda
Username   Password

  New fredmiranda.com Mobile Site
  New Feature: SMS Notification alert
  New Feature: Buy & Sell Watchlist
  

FM Forums | Alternative Gear & Lenses | Join Upload & Sell

1       2       3              7      
8
       end
  

Archive 2013 · More Mpix or perfect per-pixel quality?
  
 
AhamB
Offline
• • • • •
Upload & Sell: On
p.8 #1 · p.8 #1 · More Mpix or perfect per-pixel quality?


"diametrically opposed relative to low vs. high ISO"

wow... people who know language but not science shouldn't try to explain physics

Re. your question:

If I understand correctly, a narrow DR scene is best captured with a small well depth because you get more fine grained photon counts. From what I've understood, there is more information in the higher stops of the DR, so it's better if it's weighted towards the highlights, but if it's not you could ETTR (with the downside of increased flare, or so I've heard).



Apr 01, 2013 at 05:20 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #2 · p.8 #2 · More Mpix or perfect per-pixel quality?


Kinda reminds me of the torque/hp equation in that there is an inverse/crossover point where a given design favors one, the other or attempts to split the difference relative to the RPM curve up to the point of diminishing return ... the math involved precludes you from having it optimized for both simultaneously ... i.e. diametrically opposed/tradeoffs












Apr 01, 2013 at 05:50 PM
hiepphotog
Offline
• • •
Upload & Sell: On
p.8 #3 · p.8 #3 · More Mpix or perfect per-pixel quality?


If I understand it correctly, what Michael and Joakim are trying to say is that with a high DR scene, a deep well is preferable, while a shallow one is more suitable for a limited DR scene since it's easier to measure the difference. Since the well depth is fixed, I would expect a camera with high DR at base ISO would have a much steeper drop-off in DR when you increase the ISO.

The concept of optimizing for a certain aspect but not others is quite common in all engineering problems, though I have seen the expression "diametrically opposed." That sounds to me like an extreme polarity while in fact they might not be.



Apr 01, 2013 at 06:42 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #4 · p.8 #4 · More Mpix or perfect per-pixel quality?


Diametrically opposed ... they move in opposing directions. When one goes up, the other comes down, and the farther you move from the central (shared/crossover) point, the more extreme the variance.

The ends of a teeter-totter are diametrically opposed (as one goes up the other must come down), yet the seesaw can still be optimally balanced to the degree (variance/non-variance) that one so chooses. Even in a state of optimized balance, the ends of the seesaw are diametrically opposed.



Apr 01, 2013 at 07:19 PM
theSuede
Offline
• • • •
Upload & Sell: Off
p.8 #5 · p.8 #5 · More Mpix or perfect per-pixel quality?


Well depth isn't directly related to scene related DR (object DR), only to light sensitivity optimization. What is the lowest possible physical ISO rating you want? work from there.

Tone resolution is Poisson noise limited, it gets better as you increase the total scene light energy input. So for a maximum tonal resolution, you want very low ISO and low electronic noise.
The reason you would want smaller wells is if you rarely shoot at low ISO. A smaller well will improve high-ISO performance, and also tone resolution at high ISO since the error margins are lower for small amounts of light in a small well than for small amounts of light in a large well. Video camera sensors are typically based at ~ISO400-800, since the can't be used at longer shutter speeds than the video framerate. A smaller well does in the video case also enable a FAST readout cycle without inducing ADC speed problems.

The difference between the two DxO screenshots Rusty showed on this page is in electronic noise unrelated to FWC. The D3s has got quite noisy off-sensor ancillary electronics, meaning that when it is at low ISO the sensor is actually limited by the signal paths and the AD that's on the mainboard of the D3-type cameras (not on the sensor chip as in the Sony/Toshiba sensors). The sensor doesn't show it's best until the curve flattens down to the ideal linear loss per (ISO3200 for the D3s?). Below this, RN is limiting the DR by adding shadow noise.

But since the shadow noise is so much weaker in power than mid-gray photon shot noise - remember that we "see" in a kind of log-based curve, so 10e- of noise is A LOT in the shadows, but not even noticeable in the midtones and up - the 18% gray SNR values in the second graph is straight all the way down to ISO200 for the D3s.

Here you end up in another type of optimization, and this one takes some deep thinking to be able to grasp fully:
-"Where is the maximum of the curve you get if you plot 18% gray SNR as Y-axis vs pixel size on X-axis
---if you calculate the SNR on result images scaled to the same output size, AFTER Bayer interpolation?"

-and where is the maximum DR?




Apr 01, 2013 at 08:00 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #6 · p.8 #6 · More Mpix or perfect per-pixel quality?


theSuede wrote:

A smaller well will improve high-ISO performance, and also tone resolution at high ISO since the error margins are lower for small amounts of light in a small well than for small amounts of light in a large well.


Thanks ... this helps.


If I understand correctly:
High ISO = small well strategy
Low ISO = large well strategy

High DR Scene = small well strategy
Low DR Scene = large well strategy


Right now, all my cameras use approximately the same pixel size, targeted toward large well, low ISO approach. Up till now, I've been of the perspective that I do not need anything with high ISO performance, since I'll just use a tripod instead for low illumination levels. But, it sounds like the strategy of design for high ISO shares benefit for increased DR range.

If I've understood correctly, then a small pixel sensor could be an augmentation to my current bodies, realizing that high ISO usage may not be the only application beneficial for its use ... i.e. high DR range also.

That being said, it would seem that the more pixels debate is about more than just increased resolution, depending upon which application(s) you are shooting, i.e. high DR, low DR, shadows, highlights, @ high vs. low ISO, etc.


Thanks again,
Please continue to correct as warranted.





Apr 01, 2013 at 09:29 PM
carstenw
Offline
• • • • • •
Upload & Sell: Off
p.8 #7 · p.8 #7 · More Mpix or perfect per-pixel quality?


RustyBug wrote:
Diametrically opposed ... they move in opposing directions. When one goes up, the other comes down, and the farther you move from the central (shared/crossover) point, the more extreme the variance.

The ends of a teeter-totter are diametrically opposed (as one goes up the other must come down), yet the seesaw can still be optimally balanced to the degree (variance/non-variance) that one so chooses. Even in a state of optimized balance, the ends of the seesaw are diametrically opposed.


Unusual way to use this expression. Usually it doesn't connote motion or balance, but position. Usually, it also it used to describe irreconcilable, opposite positions, and often implies conflict.



Apr 01, 2013 at 09:48 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #8 · p.8 #8 · More Mpix or perfect per-pixel quality?


+1 @ opposing positions ... two things on opposite sides of a fulcrum point will always be opposing. Acceleration and deceleration are diametrically opposed, you can't be doing one while simultaneously doing the other, just like you can't be going up at the same time you're going down.





Apr 01, 2013 at 09:58 PM
sebboh
Offline
• • • • • •
Upload & Sell: Off
p.8 #9 · p.8 #9 · More Mpix or perfect per-pixel quality?


carstenw wrote:
Unusual way to use this expression. Usually it doesn't connote motion or balance, but position. Usually, it also it used to describe irreconcilable, opposite positions, and often implies conflict.


i believe the ends of a teeter-totter might be better described as tangentially opposed.



Apr 01, 2013 at 09:59 PM
 

Search in Used Dept. 



philip_pj
Offline
• • • •
Upload & Sell: Off
p.8 #10 · p.8 #10 · More Mpix or perfect per-pixel quality?


Presuming that DxO can accurately (or at least consistently) produce their metrics, the 24Mp sensor in the a99/RX1 and the 36Mp sensor in the D800 series appear to be so close in performance they are almost superimposed in the various charts (using RX1 to avoid the SLT hit)...both the new Sony FF sensors are also excellent at both ends of the ISO range.

Below ISO 400 the advantage of the D3s is lost, and because the D800/RX1 sensors are for almost any rational need, 'good enough' at 'high enough' ISO while being amazingly good at low to moderate ISO, the D3s looks more of a special purpose low light 'black cat in a coal mine' camera these days, with attendant shortfall of resolution.

AhamB, this argument sounds like the film days...use Velvia on drab overcast days and neg film on bright contrasty days, matching scene DR with film/sensor DR. My subjects are almost all high DR so I am a pig in mud now, and for low light work, sensor DR is radically reduced in any case, by 25-50%, a ballpark figure. In the absence of strong highlights, harmony prevails. There is always Adobe as well.



Apr 01, 2013 at 10:27 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #11 · p.8 #11 · More Mpix or perfect per-pixel quality?


sebboh wrote:
i believe the ends of a teeter-totter might be better described as tangentially opposed.


They move along the arc of the circle formed by their distance from the center/fulcrum ... where's the tangent? The length of the board is the diameter of the circle, they are at opposite ends of the diameter ... sounds diametrically opposed to me.



Apr 01, 2013 at 10:42 PM
theSuede
Offline
• • • •
Upload & Sell: Off
p.8 #12 · p.8 #12 · More Mpix or perfect per-pixel quality?


To take the thread back to the actual thread title,

-"Perfect per-pixel quality" does not, can not, and will not ever exist in a Bayer based sensor.

The absolute best we can hope for is to lower the amount of downsampling necessary to get "almost pixel perfect", and better, more efficient raw file formats.

Looking past that, we have to look at multi-layered solution like promised by either the nano-dot technologies or other new way to use more of the light that the lens sends through to the sensor.

At the moment we're only using about 12-14% of the light the lens passes. This is way to little, and here's where all the possible development areas are.



Apr 01, 2013 at 10:47 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #13 · p.8 #13 · More Mpix or perfect per-pixel quality?


philip_pj wrote:
AhamB, this argument sounds like the film days...use Velvia on drab overcast days and neg film on bright contrasty days, matching scene DR with film/sensor DR.


So, are you saying it has pragmatic application (even if not widely practiced) for technical consideration ... or it is an overkill in analysis that has little concern in today's realm. (I shot Velvia & VPS 160.)


Case in point ... I was out shooting yesterday on a partly cloudy day, the lighting would change from Sunny 16 to cloudy throughout the course of the day. Allowing for proper exposure compensation via shutter/aperture ... is there a case to be made for using a different body/pixel size or different ISO sensitivity for the variant DR of the lighting/scene?

It would be no big deal to switch bodies for targeted sensor performance than it is to switch bodies for AF performance or alt lens choice ... as long as there was rationale/reason for tangible benefit.


Edited on Apr 01, 2013 at 11:18 PM · View previous versions



Apr 01, 2013 at 10:50 PM
sebboh
Offline
• • • • • •
Upload & Sell: Off
p.8 #14 · p.8 #14 · More Mpix or perfect per-pixel quality?


RustyBug wrote:
They move along the arc of the circle formed by their distance from the center/fulcrum ... where's the tangent? The length of the board is the diameter of the circle, they are at opposite ends of the diameter ... sounds diametrically opposed to me.


the force is tangential to the circular motion. if they were each trying to pull the teeter-totter in opposite directions parallel to it's length they would be diametrically opposed.

both people in the teeter-totter are actually pushing the same direction, they just happen to have fulcrum between them.


Edited on Apr 01, 2013 at 10:57 PM · View previous versions



Apr 01, 2013 at 10:52 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #15 · p.8 #15 · More Mpix or perfect per-pixel quality?


theSuede wrote:
At the moment we're only using about 12-14% of the light the lens passes. This is way to little, and here's where all the possible development areas are.


And the other 87% of the light =



Apr 01, 2013 at 10:53 PM
mpmendenhall
Offline
• • • •
Upload & Sell: Off
p.8 #16 · p.8 #16 · More Mpix or perfect per-pixel quality?


RustyBug wrote:
And the other 87% of the light =



  1. absorbed by the Bayer color filter array
  2. hits "insensitive" parts of the sensor
  3. hits "sensitive" parts of the sensor, but fails to knock an electron loose to be counted




Apr 01, 2013 at 11:10 PM
RustyBug
Online
• • • • • •
Upload & Sell: On
p.8 #17 · p.8 #17 · More Mpix or perfect per-pixel quality?


mpmendenhall wrote:

  1. absorbed by the Bayer color filter array
  2. hits "insensitive" parts of the sensor
  3. hits "sensitive" parts of the sensor, but fails to knock an electron loose to be counted

Gotcha, thanks.



Apr 01, 2013 at 11:11 PM
1       2       3              7      
8
       end




FM Forums | Alternative Gear & Lenses | Join Upload & Sell

1       2       3              7      
8
       end
    
 

You are not logged in. Login or Register

Username   Password    Reset password