Forgot Password
Pentax Camera Forums Home
 

Reply
Show Printable Version Search this Thread
07-13-2016, 01:04 PM   #136
Pentaxian




Join Date: Nov 2013
Posts: 4,626
QuoteOriginally posted by biz-engineer Quote
I like how you turn things around. Nicolas, the only different of view point between you and clackers is that you use a theoretical model that generally works, and Clackers says that general models that generally work may fail in particular.
Honestly maybe I read it wrong but Clackers doesn't just say that in some case due to current technical implementation having bigger photosite help. He explain that this doesn't work at all and he deny practical counter examples like D810 vs K5-IIs claiming that K5-IIs has exactly the same noise as D810... That actually high iso image from D810 look practically better than K5-IIs is completely ignored because it simply doesn't fit.

The theory say he is wrong, the practice show this doesn't work like that, but still...

07-13-2016, 10:27 PM - 1 Like   #137
Loyal Site Supporter
clackers's Avatar

Join Date: Jul 2013
Location: Melbourne
Photos: Albums
Posts: 8,526
QuoteOriginally posted by Nicolas06 Quote
Man you care of 1 single photosite, you theory say if a sensor has a given single pixel of the whole sensor surface, it will be wonderfull very high quality. You signal would be like 110db and not 43, but nobody would ever interrested. You'd say look the quality is much betterbBecause your definition of the signal would be perfectly uniform color and no more....

Norm was so right about you.


The subject is noise, so don't talk instead about the tradeoff between noise and resolution. You could have 100 million tiny pixels crammed into a 1" sensor whose images post processing (the "Print Screen") couldn't rescue.


QuoteOriginally posted by Nicolas06 Quote


Honestly maybe I read it wrong but Clackers doesn't just say that in some case due to current technical implementation having bigger photosite help. He explain that this doesn't work at all and he deny practical counter examples like D810 vs K5-IIs claiming that K5-IIs has exactly the same noise as D810... That actually high iso image from D810 look practically better than K5-IIs is completely ignored because it simply doesn't fit.

The theory say he is wrong, the practice show this doesn't work like that, but still...

No, Nicolas, DxO measured the K5-IIs camera as having exactly the same noise as the D810. That's real life. That's what's in the RAW files. Then you go to a computer and try to do something about it.


DxO then guess in the Print tab at how much you might be able to improve the JPGs when you postprocess. It's a hypothetical amount that has nothing to do with sensor size or the type of image or the ISO, just the numbers of pixels involved. If you simply resized the picture, there would be no benefit at all. You have to resample as well, to do averaging.


I can say that your alarming thoughts - not supported by DxO - in this thread so far have included:



Post 97: "They will not have the same noise. The noise will be significantly lower on the FF because while the light density will be the same, the total amount of light received will be different."


Post 103: "There a relation between the sensor size and the signal ratio. If you use the same technology in both sensors, the larger sensor will get less noise because the absolute amoung of light received is greater, that is the signal is greater."


Post 106: "Because of how isos are defined, at the same iso setting, because the FF sensor is 2.3 time bigger its capacity has to be at least 2.3 time bigger than the minimal capacity required for the APSC sensor to expose correctly that sensitivity."


This is the sort of stuff that comes up in DPR threads - that's how bad it is! :-D
07-13-2016, 11:30 PM   #138
Pentaxian




Join Date: Apr 2007
Location: Sweden
Posts: 1,916
Give it up Nicolas06, clackers doesn't want to or isn't able to understand the print tab. His examples with the phone sensor or one million pixels one inch sensor show's he doesn't have a clue, even though it has been explained numerical times for him.
07-14-2016, 02:46 AM   #139
Loyal Site Supporter




Join Date: Mar 2009
Location: Southern Indiana
Photos: Gallery
Posts: 15,555
QuoteOriginally posted by clackers Quote
Norm was so right about you.


The subject is noise, so don't talk instead about the tradeoff between noise and resolution. You could have 100 million tiny pixels crammed into a 1" sensor whose images post processing (the "Print Screen") couldn't rescue.





No, Nicolas, DxO measured the K5-IIs camera as having exactly the same noise as the D810. That's real life. That's what's in the RAW files. Then you go to a computer and try to do something about it.


DxO then guess in the Print tab at how much you might be able to improve the JPGs when you postprocess. It's a hypothetical amount that has nothing to do with sensor size or the type of image or the ISO, just the numbers of pixels involved. If you simply resized the picture, there would be no benefit at all. You have to resample as well, to do averaging.


I can say that your alarming thoughts - not supported by DxO - in this thread so far have included:



Post 97: "They will not have the same noise. The noise will be significantly lower on the FF because while the light density will be the same, the total amount of light received will be different."


Post 103: "There a relation between the sensor size and the signal ratio. If you use the same technology in both sensors, the larger sensor will get less noise because the absolute amoung of light received is greater, that is the signal is greater."


Post 106: "Because of how isos are defined, at the same iso setting, because the FF sensor is 2.3 time bigger its capacity has to be at least 2.3 time bigger than the minimal capacity required for the APSC sensor to expose correctly that sensitivity."


This is the sort of stuff that comes up in DPR threads - that's how bad it is! :-D
I don't really want to continue this argument, but I just think in the real world, you see less noise in your prints when you have a bigger sensor. This is why full frame is reputed to have better high iso than APS-C. At pixel level, sure they are the same, or some full frame cameras may look even worse than APS-C cameras at this level, but you have to assume that if you have a 16 megapixel APS-C image you will either have to up size to match your 36 megapixel full frame image or down size the full frame image. Probably the fairer thing is to downsize the full frame image, in which case your noise will average out and it will look less noisy.

Anyway, I own a K3 and K5 II and a K-1 and the K-1 is better than a stop better with regard to high iso performance than either of the APS-C cameras. You can argue why (more pixels, whatever), but this ends up being true across the board, that full frames have less visible noise at high iso in prints, even if at a pixel level they have the same or even more.

07-14-2016, 03:36 AM   #140
Loyal Site Supporter
clackers's Avatar

Join Date: Jul 2013
Location: Melbourne
Photos: Albums
Posts: 8,526
QuoteOriginally posted by Gimbal Quote
Give it up Nicolas06, clackers doesn't want to or isn't able to understand the print tab. His examples with the phone sensor or one million pixels one inch sensor show's he doesn't have a clue, even though it has been explained numerical times for him.
Gimbal!

The man who once said: "So compared pixel by pixel the A7S is way ahead, it has huge pixels compared to the K50. When printed at the same size the gap closes slightly, although the A7S still wins by a great margin due to its larger total sensor area."

Here's the news: the print tab doesn't consider sensor size ... never has, never will. DxOMark explain that quite clearly.

They even present the equation, if you care to look - the link is earlier in the thread. What is the saying ... "You can lead a horse to water ..."?

The equation applies equally to phones and 1" sensors, so don't try to deny it.

Last edited by clackers; 07-14-2016 at 11:21 PM.
07-14-2016, 03:56 AM   #141
Loyal Site Supporter
clackers's Avatar

Join Date: Jul 2013
Location: Melbourne
Photos: Albums
Posts: 8,526
QuoteOriginally posted by Rondec Quote
At pixel level, sure they are the same.
Correct, Rondec.


QuoteOriginally posted by Rondec Quote
Anyway, I own a K3 and K5 II and a K-1 and the K-1 is better than a stop better with regard to high iso performance than either of the APS-C cameras.
Well, a K-1 cleans up better than a K5 II in Lightroom. If you did a resize (remove roughly three out of four pixels to get to 8Mp instead of a downsample), the performance would be the same, though, as you imply.

A Canon 5DS relies massively on downsampling from its 50Mp, because its Full Frame files are noisier than those from the Crop K5.

As I said earlier, I think anything in a signal chain - whether audio, data or photography - should be as good as possible at the start rather than hope that filters and boosters will do it all further down the line - they have costs of their own.

The bigger pixels in the K-1 give a native advantage in both noise and dynamic range over the K-3.

It's why I bought one, not the other.

Last edited by clackers; 07-14-2016 at 04:06 AM.
07-14-2016, 05:44 AM   #142
Pentaxian




Join Date: Feb 2015
Photos: Albums
Posts: 3,245
Noise is only one element. I'd say that if you don't consider diffraction, in theory, it is more beneficial to have more smaller pixels because downsizing also improves the dymanic range when summing up pixels the random noise is also averaged out as well as the resolution. But smaller pixels diffract, especially the light gathering capability of the red component is more sensitive to pixel size, so, once you reach 2um pixel size (like it is for compact cameras or phones), the image quality drops significantly. There seems to be a sweet spot that's around 5um pixel pitch, low diffraction sensor array. If you cram 28Mpixel in a K3 format (like samsung did), you don't get any better IQ because you get into sampling additional higher frequencies having lower contrast from the lens and the per pixel noise increases as well. Pentax/Ricoh decision to introduce the K1 36Mp after the K3 simply makes sense. Ricoh upgraded their apsc camera line sensor resolution until they reached a point of diminishing returns with the K3. Before the K3, you did not need to go full frame to get more IQ. K1 sensor pixel pitch (like K5IIs) is sort of a sweet spot, because OAA is not mandatory and pixels are large enough to get decent light absorption. While K1, D810 are seating on the top of the IQ curve, Canon 5Ds hits the downward curve in terms of DR yet maximize the resolution that can be squeeze out of a good lens without too much risk of moiré.

I'm referring to "Full Sensor Apparent Image Quality (FSAIQ) from here Clarkvision: Digital Camera Review and Sensor Performance Summary

And for normhead (hiking and shooting wildlife on a budget without carrying tons), I'd refer to " Focal Length Limited Apparent Image Quality FLL-AIQ-MAX" here http://www.clarkvision.com/articles/digital.sensor.performance.summary/#FLLAIQMAX

Last edited by biz-engineer; 07-14-2016 at 05:58 AM.
07-14-2016, 07:16 AM   #143
Pentaxian
normhead's Avatar

Join Date: Jun 2007
Location: Near Algonquin Park
Photos: Gallery | Albums
Posts: 24,956
QuoteOriginally posted by biz-engineer Quote
Noise is only one element. I'd say that if you don't consider diffraction, in theory, it is more beneficial to have more smaller pixels because downsizing also improves the dymanic range when summing up pixels the random noise is also averaged out as well as the resolution. But smaller pixels diffract, especially the light gathering capability of the red component is more sensitive to pixel size, so, once you reach 2um pixel size (like it is for compact cameras or phones), the image quality drops significantly. There seems to be a sweet spot that's around 5um pixel pitch, low diffraction sensor array. If you cram 28Mpixel in a K3 format (like samsung did), you don't get any better IQ because you get into sampling additional higher frequencies having lower contrast from the lens and the per pixel noise increases as well. Pentax/Ricoh decision to introduce the K1 36Mp after the K3 simply makes sense. Ricoh upgraded their apsc camera line sensor resolution until they reached a point of diminishing returns with the K3. Before the K3, you did not need to go full frame to get more IQ. K1 sensor pixel pitch (like K5IIs) is sort of a sweet spot, because OAA is not mandatory and pixels are large enough to get decent light absorption. While K1, D810 are seating on the top of the IQ curve, Canon 5Ds hits the downward curve in terms of DR yet maximize the resolution that can be squeeze out of a good lens without too much risk of moiré.

I'm referring to "Full Sensor Apparent Image Quality (FSAIQ) from here Clarkvision: Digital Camera Review and Sensor Performance Summary

And for normhead (hiking and shooting wildlife on a budget without carrying tons), I'd refer to " Focal Length Limited Apparent Image Quality FLL-AIQ-MAX" here Clarkvision: Digital Camera Review and Sensor Performance Summary
I should bookmark that.

07-14-2016, 11:50 AM   #144
Pentaxian




Join Date: Apr 2007
Location: Sweden
Posts: 1,916
QuoteOriginally posted by clackers Quote
Gimbal!

The man who once said: "So compared pixel by pixel the A7S is way ahead, it has huge pixels compared to the K50. When printed at the same size the gap closes slightly, although the A7S still wins by a great margin due to its larger total sensor area."

Here's the news: the print tab doesn't consider sensor size ... never has, never will. DxOMark explain that quite clearly. They even present the equation, if you care to look - the link is earlier in the thread. What is the saying ... "You can lead a horse to water ..."?

The equation applies equally to phones and 1" sensors, so don't try to deny it.
I wrote that a long time ago, and it's still true. I rest my case, as they say.
07-14-2016, 11:17 PM   #145
Loyal Site Supporter
clackers's Avatar

Join Date: Jul 2013
Location: Melbourne
Photos: Albums
Posts: 8,526
QuoteOriginally posted by Gimbal Quote
I wrote that a long time ago, and it's still true.
Wow. I have nothing more to add ... good night!
07-15-2016, 12:24 AM   #146
Pentaxian




Join Date: Feb 2015
Photos: Albums
Posts: 3,245
Down-sampling = taking one pixel out of a number of pixels. Usually, resizing involves adding up pixels, due to random nature of noise which may be a positive luminance error of one cell and a negative error of luminance in the adjacent cell, when adding up two cell values, some of the luminance error is cancelled out and the relative amount of useful signal remains the same. Normally, resizing reduces the noise level in an image, as well as the amount of information contained in the image. In theory (model of DXO), the noise is purely random, so when resizing to 8Mpixels the DXO model improves the SNR systematically regardless of the nature of the noise (considering the noise is purely random for all sensors). Practically.. when the noise comes in clusters due to a sensor substrate event around a certain location (for instance a spike generated in the substrate by a read operation), the luminance errors in the neighborhood of a cell are correlated (for example, all errors are positive increase of luminance for a group of pixels), averaging does not cancel out the luminance error but the error remains mixed with the wanted signal. That's why, if you resize a D810 image to 12Mpixels, in theory, you should get less noise than the image from the A7s, but practically, the A7s image has less noise. The reason why some of the noise is correlated, is that all pixels seat on a common substrate and noise propagates in the same way as heat. You can see this effect on low light images, you can see noise patches which affect small groups of pixels. The DXO model of print SNR is optimistic.
07-15-2016, 01:53 AM   #147
Pentaxian
kh1234567890's Avatar

Join Date: Sep 2010
Location: Manchester, UK
Photos: Gallery
Posts: 2,386
QuoteOriginally posted by biz-engineer Quote
Practically.. when the noise comes in clusters due to a sensor substrate event around a certain location (for instance a spike generated in the substrate by a read operation), the luminance errors in the neighborhood of a cell are correlated (for example, all errors are positive increase of luminance for a group of pixels), averaging does not cancel out the luminance error but the error remains mixed with the wanted signal
A perfectly valid point, IMO. As with all the numerology the numbers are meaningless without quoting the error bounds of their method.

Really, they should go through the whole process - demosaicing by a documented algorithm, gamma mapping, JPEG compression mangling, JPEG decompression, rescaling and further colour mapping by the minilab printing machine and a reflectance measurement off the finished print under given illumination.
07-15-2016, 02:16 AM   #148
Loyal Site Supporter




Join Date: Mar 2009
Location: Southern Indiana
Photos: Gallery
Posts: 15,555
QuoteOriginally posted by biz-engineer Quote
Down-sampling = taking one pixel out of a number of pixels. Usually, resizing involves adding up pixels, due to random nature of noise which may be a positive luminance error of one cell and a negative error of luminance in the adjacent cell, when adding up two cell values, some of the luminance error is cancelled out and the relative amount of useful signal remains the same. Normally, resizing reduces the noise level in an image, as well as the amount of information contained in the image. In theory (model of DXO), the noise is purely random, so when resizing to 8Mpixels the DXO model improves the SNR systematically regardless of the nature of the noise (considering the noise is purely random for all sensors). Practically.. when the noise comes in clusters due to a sensor substrate event around a certain location (for instance a spike generated in the substrate by a read operation), the luminance errors in the neighborhood of a cell are correlated (for example, all errors are positive increase of luminance for a group of pixels), averaging does not cancel out the luminance error but the error remains mixed with the wanted signal. That's why, if you resize a D810 image to 12Mpixels, in theory, you should get less noise than the image from the A7s, but practically, the A7s image has less noise. The reason why some of the noise is correlated, is that all pixels seat on a common substrate and noise propagates in the same way as heat. You can see this effect on low light images, you can see noise patches which affect small groups of pixels. The DXO model of print SNR is optimistic.
The thing is that if you look at the DXO Mark graphs and compare the D810 to the A7s, they are very close with regard to noise, but the A7s is a lot better with regard to dynamic range after iso 3200. To me, this is what allows the A7s images to look better at high iso regardless of the presence or absence of noise. While we haven't mentioned it in this discussion, when you push iso up, your dynamic range goes down and cameras that can maintain more dynamic range are going to look better than those that do not, even if they have just as much noise.
07-15-2016, 05:46 AM   #149
Pentaxian
normhead's Avatar

Join Date: Jun 2007
Location: Near Algonquin Park
Photos: Gallery | Albums
Posts: 24,956
QuoteOriginally posted by Rondec Quote
The thing is that if you look at the DXO Mark graphs and compare the D810 to the A7s, they are very close with regard to noise, but the A7s is a lot better with regard to dynamic range after iso 3200. To me, this is what allows the A7s images to look better at high iso regardless of the presence or absence of noise. While we haven't mentioned it in this discussion, when you push iso up, your dynamic range goes down and cameras that can maintain more dynamic range are going to look better than those that do not, even if they have just as much noise.
And that's a point the people concerned about "what camera is better at high ISO" tend to ignore. As you're ISO goes up, you Dynamic range goes down. To a certain extent this in compensated for by the fact that in low light there is less contrast in the scene, so you need less dynamic range to capture it. and wishing reason, loss of Dynamic Ranges is correctable in post processing. But in many images it's not the noise that ruins your high ISO image, it's the loss of dynamic range, and uncontrolled light sources creating shadows in odd places.
07-15-2016, 12:13 PM   #150
Pentaxian




Join Date: Nov 2013
Posts: 4,626
QuoteOriginally posted by kh1234567890 Quote
A perfectly valid point, IMO. As with all the numerology the numbers are meaningless without quoting the error bounds of their method.

Really, they should go through the whole process - demosaicing by a documented algorithm, gamma mapping, JPEG compression mangling, JPEG decompression, rescaling and further colour mapping by the minilab printing machine and a reflectance measurement off the finished print under given illumination.
Yeah that why there so much criticism here. Everybody got it right at home with proof that DxO did it wrong. That soooo bad.
Reply

Bookmarks
  • Submit Thread to Facebook Facebook
  • Submit Thread to Twitter Twitter
  • Submit Thread to Digg Digg
Tags - Make this thread easier to find by adding keywords to it!
aperture, aps-c, background, blur, camera, dof, dslr, feet, ff, ff over aps-c, field, focus, frame, full frame, full-frame, k-1, k1, lens, paper, pentax k-1, people, picture, posters, stitching
Thread Tools Search this Thread
Search this Thread:

Advanced Search


Similar Threads
Thread Thread Starter Forum Replies Last Post
FF vs APS-C Field of View revisited Ole Pentax K-1 2 05-07-2016 02:13 PM
Confused about Angle of View of Lenses on FF vs APS-C? Kath Pentax SLR Lens Discussion 3 10-01-2015 09:55 AM
Does FF vs APS-C affect amount of light? windhorse General Photography 46 03-02-2015 07:07 PM
Quick question regarding field of view - FF vs APS-C glass? Julie Pentax SLR Lens Discussion 5 12-23-2012 05:33 PM
APS-C does not increase focal length over FF, it decreases field of view. TomTextura Photographic Technique 135 06-09-2012 04:58 PM



All times are GMT -7. The time now is 09:45 AM. | See also: NikonForums.com, CanonForums.com part of our network of photo forums!
  • Red (Default)
  • Green
  • Gray
  • Dark
  • Dark Yellow
  • Dark Blue
  • Old Red
  • Old Green
  • Old Gray
  • Dial-Up Style
Hello! It's great to see you back on the forum! Have you considered joining the community?
register
Creating a FREE ACCOUNT takes under a minute, removes ads, and lets you post! [Dismiss]
Top