Pages: 1 [2] 3   Go Down

Author Topic: End of the megapixel race? Two interesting articles by Ctein  (Read 20163 times)

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8831
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #20 on: November 29, 2014, 06:11:06 pm »

With a highly dense sensor you are not capturing real detail in large DOF applications, just blurred redundant pixels.

Hi Guillermo,

Yes, but oversampling the diffraction blur will allow to restore a large part of it with deconvolution sharpening.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

Guillermo Luijk

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1882
    • http://www.guillermoluijk.com
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #21 on: November 29, 2014, 06:21:03 pm »

Bart, do you think the use and good performance of deconvolution techniques will become of general use for this kind of applications? or users will rather be happy with tons of blurred Mpx thinking they are gaining something over lower pixel count sensors?
« Last Edit: November 29, 2014, 06:22:44 pm by Guillermo Luijk »
Logged

dwswager

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1375
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #22 on: November 29, 2014, 08:29:47 pm »

Well... people who think 36MP is 3 times the resolution (3 times the amount of pixels to be exact) of 12MP are correct, no argument around it.

Pixel count is NOT a resolution, but a proxy developed to simplify for various aspect ratios of images.  A 36 MP camera of the same aspect ratio  as a 12 MP camera will not generate an image that is twice as large even though it is composed of 24 million more pixels.

Assuming a 3x2 aspect ratio of specific dimension:

Nikon D300, 12.3MP, 4,288 x 2,848
Nikon D810, 36.3MP, 7,360 x 4,912  (Not twice 4288 x 2,848)

8576 x 5696 (48.8MP) is twice 4,288 x 2,848 (12.3MP).  This is why discussing the difference between 12MP and 16MP is rather pointless because they are basically the same thing.
Logged

Geraldo Garcia

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 470
    • Personal blog
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #23 on: November 29, 2014, 08:52:43 pm »

I consider myself a technical minded person and I love to make this kind of analysis and calculations, but there is one thing that we can't argue with: the reality.

As a commercial printer specialized on art I print for many artists (photographers among them) and I see a lot of files from a lot of cameras, so I can tel you this by personal experience:
Pick a good sharp photo properly taken with a 12MP FF camera.
Pick another good sharp photo properly taken with a 36MP FF camera.
Now print both photos 36x54" and see the difference. The 36MP image will print way better

There is no real argument after you do this. Honestly.
We can speculate about greater MP count, but at the current level the superiority of the 36MP is brutal even with diffraction and other optical constrains.
Logged

shadowblade

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2707
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #24 on: November 29, 2014, 09:56:54 pm »

Pixel count is NOT a resolution, but a proxy developed to simplify for various aspect ratios of images.  A 36 MP camera of the same aspect ratio  as a 12 MP camera will not generate an image that is twice as large even though it is composed of 24 million more pixels.

Assuming a 3x2 aspect ratio of specific dimension:

Nikon D300, 12.3MP, 4,288 x 2,848
Nikon D810, 36.3MP, 7,360 x 4,912  (Not twice 4288 x 2,848)

8576 x 5696 (48.8MP) is twice 4,288 x 2,848 (12.3MP).  This is why discussing the difference between 12MP and 16MP is rather pointless because they are basically the same thing.


Nothing to do with aspect ratios at all.

Basically, pixel counts increase as a square of linear resolution. To double the linear resolution of a 12MP file, you need 48MP. To double the linear resolution of 9MP, you need 36MP. Going from 22 to 36MP represents a not-insignificant 30% or so increase in linear resolution.

This is because doubling the size in both length and width means you need four times as many pixels. And that's why high pixel count is more relevant than ever - 80MP may sound like a huge pixel count, but only represents double the linear resolution of a 20MP image. If you want high-quality prints at a large size, you can always use the resolution.
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11296
    • Echophoto
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #25 on: November 30, 2014, 03:04:32 am »

Hi,

I think so, too.

May I ask two questions?

When I went from 12 MP to 24 MP I could see some advantage to 24 MP in A2-size prints. I see very little advantage going from 24 MP to 39 MP at A2 (16x23"). So I feel that 12 MP are pretty much OK for 16x23" and 16MP or so may be really good for that print size. What is your take on that?

Reason for asking: It is about the size of prints desktop printers can make, and it is about 16MP 4/3 cameras can produce. Ctein said in an interview on LuLa that 16 MP from APS-C is good enough for A2.

The other question is, do you see benefits going to higher resolution on full frame (24x36)?

Reason for asking: Some guys (like Jim Kasson, myself and others) have done some research and came to the conclusion that small pixels will give better reproduction. It would be interesting to hear your view. Is the Nikon D810 the summit, or just a plateau on the way to a summit far away?

Best regards
Erik

I consider myself a technical minded person and I love to make this kind of analysis and calculations, but there is one thing that we can't argue with: the reality.

As a commercial printer specialized on art I print for many artists (photographers among them) and I see a lot of files from a lot of cameras, so I can tel you this by personal experience:
Pick a good sharp photo properly taken with a 12MP FF camera.
Pick another good sharp photo properly taken with a 36MP FF camera.
Now print both photos 36x54" and see the difference. The 36MP image will print way better

There is no real argument after you do this. Honestly.
We can speculate about greater MP count, but at the current level the superiority of the 36MP is brutal even with diffraction and other optical constrains.
« Last Edit: November 30, 2014, 03:07:20 am by ErikKaffehr »
Logged
Erik Kaffehr
 

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11296
    • Echophoto
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #26 on: November 30, 2014, 03:20:01 am »

Hi,

My take is that having a blurred image that is properly sharpened and downsized to say half of the resolution would yield a superior rendition compared with an image having half of the resolution to start with.

I have a nifty pair of images showing it, but would I post it would cause immediate retaliation from the self appointed forum cop, I want to avoid it.
http://echophoto.dnsalias.net/ekr/index.php/photoarticles/78-aliasing-and-supersampling-why-small-pixels-are-good

Another point is that I don't think we shoot f/16 for landscape photography. Personally I mostly shoot f/8 stopping down when needed. On the Hasselblad my standard aperture is f/11. I also asked what aperture folks use on MFD and it seems that most users would shoot f/11, but technical camera users were often using f/5.6 (using technical lenses on technical cameras).

It seems that good decent restoration is possible from f/16 images. Clearly, that needs deconvolution, but we have deconvolution in Lightroom and there is FocusMagic that is easy to use and gives good results.

Best regards
Erik

Bart, do you think the use and good performance of deconvolution techniques will become of general use for this kind of applications? or users will rather be happy with tons of blurred Mpx thinking they are gaining something over lower pixel count sensors?

« Last Edit: November 30, 2014, 03:31:11 am by ErikKaffehr »
Logged
Erik Kaffehr
 

Manoli

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1901
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #27 on: November 30, 2014, 04:40:55 am »

so I can tel you this by personal experience:
Pick a good sharp photo properly taken with a 12MP FF camera.
Pick another good sharp photo properly taken with a 36MP FF camera.
Now print both photos 36x54" and see the difference. The 36MP image will print way better

NO great surprise here.
The 36MP image will print 54" ( long end) at 136 ppi - native (7300x4900)
The 12MP image will print 54" at 74 ppi - native (4000x3000)

ppi figures before interpolation.

Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8831
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #28 on: November 30, 2014, 05:02:27 am »

Bart, do you think the use and good performance of deconvolution techniques will become of general use for this kind of applications? or users will rather be happy with tons of blurred Mpx thinking they are gaining something over lower pixel count sensors?

Guillermo, I see several Raw converters already having (limited) deconvolution capabilities. Not all are implemented as good as what is possible, but that's also a bit a chicken and egg situation. If we had hugely oversampling sensor designs, it would probably be implemented in a camera ASIC and firmware.

RawTherapee, and Iridient Raw Developer have Richardson-Lucy deconvolution built-in as one of the sharpening options. Adobe's ARC and Lightroom do some sort of deconvolution when the Detail slider control is pushed towards 100, and Photoshop has SmartSharpening which is deconvolution based in Advanced mode.

There are dedicated deconvolution solutions from FocusMagic, Topaz Infocus, Piccure+, and various Astrophotography oriented image processors. There is lots of ongoing research which shows the tremendous capabilities of this approach if implemented well, some even based on a simple single planoconvex lens element (also discussed here).

Whether lens blur or oversampling blur, both can be restored to the original input signal to a large extent.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

dwswager

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1375
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #29 on: November 30, 2014, 12:20:59 pm »

Nothing to do with aspect ratios at all.

Resolution no, but pixel count yes.  The only way a MP figure makes any sense at all is when the aspect is known.  Marketing guys can use it because they are comparing 2 things with the same or very similar aspect ratios.  Within brand it is useful because even with different sized sensors, they have the same aspect.

Interestingly, my 1st exposure to MP came with my 1st digital printing experience of a film scan.  The printer asked the MP count of the image.  Of course, he knew the printing resolution and the aspect ratio of the output (and believe he assumed same aspect of the input file) so all he needed was MP count to make a general statement concerning the potential quality of the output.
Logged

Geraldo Garcia

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 470
    • Personal blog
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #30 on: November 30, 2014, 01:51:31 pm »

NO great surprise here.
The 36MP image will print 54" ( long end) at 136 ppi - native (7300x4900)
The 12MP image will print 54" at 74 ppi - native (4000x3000)
ppi figures before interpolation.

Sure no great surprise to you or to me, but I was directing my thought to the argument about diffraction limiting a f16 36MP image to 12MP level of detail.
I am sure diffraction and other optical problems do play an important role, but not enough to invalidate the gains of a higher MP sensor. Not at the current level at least.
Logged

LKaven

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1060
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #31 on: November 30, 2014, 02:13:00 pm »

It's worth remembering that linear resolution and pixel count are both useful ways to measure (commonsense) resolution.  Going from 12MP to 24MP allows one to print twice the area at the same quality.  That is a useful measure for some purposes.

Eric Fossum thinks that the bayer sensor will top out around 200MP on full-frame.  As far as I'm concerned, this is a good thing, and one can always decide later how much the total information available they want or need. 

Beyond that though is (we hope) Fossum's QIS (Quantum Image Sensor) with its single-photon photosites and temporal processing.  It helps to remember the increasing role that temporal integration will play in the future.  All of this of course happens at the sub-diffraction level.  Nothing wrong with that either.  There is plenty of /information/ at that level, that can be exploited. 

In the end, we must remember that as the technology progresses, the assumptions do not stay fixed, and there is a significant qualitative and conceptual shift to be expected.

Geraldo Garcia

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 470
    • Personal blog
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #32 on: November 30, 2014, 06:08:04 pm »

May I ask two questions?

When I went from 12 MP to 24 MP I could see some advantage to 24 MP in A2-size prints. I see very little advantage going from 24 MP to 39 MP at A2 (16x23"). So I feel that 12 MP are pretty much OK for 16x23" and 16MP or so may be really good for that print size. What is your take on that?

Reason for asking: It is about the size of prints desktop printers can make, and it is about 16MP 4/3 cameras can produce. Ctein said in an interview on LuLa that 16 MP from APS-C is good enough for A2.

The other question is, do you see benefits going to higher resolution on full frame (24x36)?

Reason for asking: Some guys (like Jim Kasson, myself and others) have done some research and came to the conclusion that small pixels will give better reproduction. It would be interesting to hear your view. Is the Nikon D810 the summit, or just a plateau on the way to a summit far away?

Hello Erik,

I do agree with everything you said, we just have to remember that "good enough" does not mean "best possible". I am sure that we would be able to see improvements on an A2 print from a 36MP file on a side by side comparison against another from a 16MP file... if we print both on a smooth glossy paper and look closely. On a matte textured paper the differences would be less noticeable or negligible. Without the side by side comparison I definitely agree the 16MP would be more than good enough.

About going to higher resolutions on FF (24x36mm), I can only guess, but I think we will see benefits. First because I tend to believe that we can still extract more "juice" of our current lenses. Surely optical problems can make our 36MP camera capture only a level of detail equivalent to, lets say, 24MP. But that same lens on an hypothetical 72MP camera may be able to render, lets say, 48MP level of detail. I think the limitations are not like a brick wall but more like a gear problem that makes you car loose a percentage of the power.

Besides, as you said, there is the software side of the coin that may be able to extract more detail from a larger amount of data (even if apparently redundant). I am still a bit skeptical about using deconvolution sharpening for other forms recovery that are not related to motion, I believe that it does not work consistently varying from image to image. But, admittedly, I have not tested it properly.

At this moment, as I print large images for living, I want more resolution for sure... until the day I could not see a difference on a 44x66" print!  ;D

BTW, last week I received some files from an IQ180. Studio shots, portraits, properly lighted, camera on a tripod, Schneider glass... The client ordered some A1 prints but I gave him a free 44x58" just because I wanted to see that file printed on that size. It was surely the best print on that size (from a single click) that I ever saw, but honestly I saw some room for improvement.

Regards.
« Last Edit: November 30, 2014, 06:11:39 pm by Geraldo Garcia »
Logged

MoreOrLess

  • Full Member
  • ***
  • Offline Offline
  • Posts: 239
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #33 on: December 01, 2014, 12:37:59 am »

As far as the old "nature of resolution" argument goes it always seemed a bit pointless to me simply because most people are likely to have the same feelings for input AND output. That is someone who thinks 24 MP is double 12 MP is resolution thinks doubling the print output means doubling the area.

I would argue the latter is in the realms of most peoples likely use as well, you start doubling the dimensions of each side of the print and you very quickly get to sizes beyond those printed by even many professional landscape shooters.
Logged

dwswager

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1375
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #34 on: December 01, 2014, 03:48:53 pm »

As far as the old "nature of resolution" argument goes it always seemed a bit pointless to me simply because most people are likely to have the same feelings for input AND output. That is someone who thinks 24 MP is double 12 MP is resolution thinks doubling the print output means doubling the area.

I would argue the latter is in the realms of most peoples likely use as well, you start doubling the dimensions of each side of the print and you very quickly get to sizes beyond those printed by even many professional landscape shooters.

When was the last time anyone ever came up to you and said "I just took the most spectacular image and I'm having it printed at 187 SQIN to hang in my living room!"

The fact of the matter is that an image has a set aspect ratio and therefore as one dimension changes, so does the other.  I know for certain that 16x20 is double 8x10 while the actual printed area is 4 times as large!  In addition, real world objects are represented in 2 dimensions in an image.  If I print an 8x10 where an object is reproduced at 1/2 life size, then to get to life size, I have to double the size of the print...16 x 20. 

It is not that MP count and knowing that double the MP means a doubling of the print area is useless, but it can be misleading.  95% of people that own cameras believe they can make a print twice as large from a camera with double the megapixels and they aren't thinking I can print twice the area.  Monitor and TV people got around this by specifying an aspect and diagonal length.
Logged

SZRitter

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 384
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #35 on: December 01, 2014, 04:03:30 pm »

When was the last time anyone ever came up to you and said "I just took the most spectacular image and I'm having it printed at 187 SQIN to hang in my living room!"

The fact of the matter is that an image has a set aspect ratio and therefore as one dimension changes, so does the other.  I know for certain that 16x20 is double 8x10 while the actual printed area is 4 times as large!  In addition, real world objects are represented in 2 dimensions in an image.  If I print an 8x10 where an object is reproduced at 1/2 life size, then to get to life size, I have to double the size of the print...16 x 20. 

It is not that MP count and knowing that double the MP means a doubling of the print area is useless, but it can be misleading.  95% of people that own cameras believe they can make a print twice as large from a camera with double the megapixels and they aren't thinking I can print twice the area.  Monitor and TV people got around this by specifying an aspect and diagonal length.

It is all perspective. The double of 20 sq inches is 40 sq inches. But, the double of 4" x 5" (aka, 20sq inches) is an 8" x 10" (aka, 80sq inches). I for one never saw pixels as the same thing as print size, but rather the count of individual pixels in a given area. So yes, 16 MegaPixels is twice as much as 8 MegaPixels, as the unit in question is the individual pixels, not the pixel dimensions.

As for the plateau, as someone who uses a 36MP camera phone, let me tell you, I am a fan of oversampling. We have not hit it yet, and there is more to come. But, we will need faster electronics and larger storage to make it a feasible endeavor. I would think that a sensor will have hit it's max when it can out resolve the best glass at it's best settings. Once we hit that point, although we will not have a technological plateau, we will have a plateau in the in the optimal mixture of sensor resolution vs lens.
Logged

LKaven

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1060
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #36 on: December 01, 2014, 04:24:12 pm »

A /lot/ of people, especially those with printers having 17" platens decide between printing 8x10, or 11x17, two popular paper sizes at reasonable prices.  It's a meaningful difference, and (within 1.0 smidgen), a doubling in print area.  A 16x20 is a splurge.

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11296
    • Echophoto
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #37 on: December 01, 2014, 04:27:47 pm »

Geraldo,

Thanks a lot for your observations!

Best regards
Erik
Logged
Erik Kaffehr
 

dwswager

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1375
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #38 on: December 01, 2014, 04:45:33 pm »

A /lot/ of people, especially those with printers having 17" platens decide between printing 8x10, or 11x17, two popular paper sizes at reasonable prices.  It's a meaningful difference, and (within 1.0 smidgen), a doubling in print area.  A 16x20 is a splurge.

Extreme money saving tip.  If your are deciding between 8x10 or 11x17, get a printer with a 13" platen and save a couple grand!  It will even print 13x19 in borderless mode.  As someone with a Epson printer with a 13" platen, I fantasize about ways to sneak a printer with a 17" wide platen into my home office without the SuperWife suspecting precisely because I want to print 16 x 20 and larger multi-shot (not really panoramic) images!
Logged

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2370
    • The Last Word
Re: End of the megapixel race? Two interesting articles by Ctein
« Reply #39 on: December 01, 2014, 04:51:31 pm »

Extreme money saving tip.  If your are deciding between 8x10 or 11x17, get a printer with a 13" platen and save a couple grand!  It will even print 13x19 in borderless mode.  As someone with a Epson printer with a 13" platen, I fantasize about ways to sneak a printer with a 17" wide platen into my home office without the SuperWife suspecting precisely because I want to print 16 x 20 and larger multi-shot (not really panoramic) images!

In the case of every printer I've ever owned, I've spent far more money on ink than on the printer itself. Larger printers tend to have bigger ink carts and lower cost per ml of ink. So, the smaller printer isn't always a bargain.

Jim
Pages: 1 [2] 3   Go Up