Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: The superiority of the 20D pixel pitch  (Read 12872 times)
Ray
Sr. Member
****
Offline Offline

Posts: 8908


« on: July 27, 2006, 01:06:51 AM »
ReplyReply

My first digital camera was the D60 with a pixel pitch close to that of the current 1Ds2. Shortly after the D60 was released, we had the 1Ds with a larger number of pixels and a larger sensor.

I always though it curious that 1Ds owners would emphatically declare, on this site and elsewhere, that their camera was 'ruthless' with inferior quality lenses and that to get the best from the camera you needed really good lenses. In other words, lenses are the weak link.

I found it curious because my common sense (nous) told me that all  cameras perform better with better lenses. The 1Ds or 1Ds2 is no exception.

Now that I'm an owner of a 5D, I can test this 'common sense' theory of mine.

I first carried out tests with a rather average zoom, the 100-400L IS, with and without 1.4x extender, and demonstrated some surprising results.

Suspecting the results might be a bit inconclusive because of the inferiority of the lens (which was actually described as crap by a person whose name I shall not mention), I decided to repeat the test with a better quality lens at a close distance to a test chart of lines and textures, which I constructed myself for the purpose.

My sharpest lens is the humble Canon 50/1.8, which I used for the test. The next sharpest would be the much more expensive TS-E 90mm.

The following images are a bit large (around 2MB after download) but they demonstrate clearly that even at f22, the 20D delivers more detail than the 5D, after cropping the 5D image to the same size as the 20D image then interpolating.

There's always an issue of sharpening. Should one sharpen the interpolated 5D image and not the uninterpolated 20D image? Should one sharpen both images, but the interpolated 5D image more?

I've chosen the latter option as a compromise with concerns of Bill Janes, althought it's clear to me that sharpening does not alter the outcome. The 5D images are sharpened more than the 20D images in the following examples.

For the benifit of those who don't want to bother downloading the test images, the results are unambiguous. The 20D produces more image detail and better contrast at both f22 and f8, than the 5D after cropping.

This is good news for those who are expecting a 22mp 5D or 1Ds2 successor.

The first image is an overview of my test chart which consists of Norman Koren line charts, fluffy twine and various grades of sandpaper pasted to a plywood base with wood grain. After viewing a TV documantary on Andy Warwhol (have I spelt that correctly?) it did occur to me for a brief fraction of a second that perhaps I should urinate on my carefully constructed test chart to create some subtle gradients. I decided against it on the grounds of hygiene.

[attachment=842:attachment]

The following 3 images are really self explanatory.

[attachment=841:attachment]   [attachment=843:attachment]   [attachment=844:attachment]
Logged
BernardLanguillier
Sr. Member
****
Offline Offline

Posts: 8208



WWW
« Reply #1 on: July 27, 2006, 05:00:42 AM »
ReplyReply

Quote
After viewing a TV documantary on Andy Warwhol (have I spelt that correctly?) it did occur to me for a brief fraction of a second that perhaps I should urinate on my carefully constructed test chart to create some subtle gradients. I decided against it on the grounds of hygiene.
[a href=\"index.php?act=findpost&pid=71820\"][{POST_SNAPBACK}][/a]

That was probably the right decision.

Those guys complaining about how demanding the 1ds was on their lenses were probably speaking about the corners... and your test kind of supports their statement since you reached the conclusion that the lenses you tested had no problem dealing with the extra detail prodived by the 20D as a result of the higher pixel density [in the APS- sized center section of the image circle].

I am not sure that your conclusion about a 22MP 1ds3 is a logical deduction of your test results, since you are overlooking how poorly images would look in the corners of images taken with current wide angle lenses with such high pixel count FF sensors...

Cheers,
Bernard
Logged

A few images online here!
francois
Sr. Member
****
Offline Offline

Posts: 6877


« Reply #2 on: July 27, 2006, 05:17:35 AM »
ReplyReply

Quote
...
After viewing a TV documantary on Andy Warwhol (have I spelt that correctly?) it did occur to me for a brief fraction of a second that perhaps I should urinate on my carefully constructed test chart to create some subtle gradients. I decided against it on the grounds of hygiene
...
Ray,
Your 5D is NOT weather-sealed!  
Logged

Francois
bjanes
Sr. Member
****
Online Online

Posts: 2825



« Reply #3 on: July 27, 2006, 09:44:56 AM »
ReplyReply

Quote
My first digital camera was the D60 with a pixel pitch close to that of the current 1Ds2. Shortly after the D60 was released, we had the 1Ds with a larger number of pixels and a larger sensor.

I always though it curious that 1Ds owners would emphatically declare, on this site and elsewhere, that their camera was 'ruthless' with inferior quality lenses and that to get the best from the camera you needed really good lenses. In other words, lenses are the weak link.

I found it curious because my common sense (nous) told me that all  cameras perform better with better lenses. The 1Ds or 1Ds2 is no exception.

Now that I'm an owner of a 5D, I can test this 'common sense' theory of mine.

I first carried out tests with a rather average zoom, the 100-400L IS, with and without 1.4x extender, and demonstrated some surprising results.

Suspecting the results might be a bit inconclusive because of the inferiority of the lens (which was actually described as crap by a person whose name I shall not mention), I decided to repeat the test with a better quality lens at a close distance to a test chart of lines and textures, which I constructed myself for the purpose.

My sharpest lens is the humble Canon 50/1.8, which I used for the test. The next sharpest would be the much more expensive TS-E 90mm.

The following images are a bit large (around 2MB after download) but they demonstrate clearly that even at f22, the 20D delivers more detail than the 5D, after cropping the 5D image to the same size as the 20D image then interpolating.

There's always an issue of sharpening. Should one sharpen the interpolated 5D image and not the uninterpolated 20D image? Should one sharpen both images, but the interpolated 5D image more?

I've chosen the latter option as a compromise with concerns of Bill Janes, althought it's clear to me that sharpening does not alter the outcome. The 5D images are sharpened more than the 20D images in the following examples.

For the benifit of those who don't want to bother downloading the test images, the results are unambiguous. The 20D produces more image detail and better contrast at both f22 and f8, than the 5D after cropping.

This is good news for those who are expecting a 22mp 5D or 1Ds2 successor.

The first image is an overview of my test chart which consists of Norman Koren line charts, fluffy twine and various grades of sandpaper pasted to a plywood base with wood grain. After viewing a TV documantary on Andy Warwhol (have I spelt that correctly?) it did occur to me for a brief fraction of a second that perhaps I should urinate on my carefully constructed test chart to create some subtle gradients. I decided against it on the grounds of hygiene.


[{POST_SNAPBACK}][/a]

Ray,

Your tests are interesting and demonstrate a fundamental point about MTF. Photographers who say that the camera outresolves the lens are speaking rather simplistically, and appear to be assuming that the imaging chain is no stronger than its weakest link.

In a test such as yours, the system MTF is being tested and this includes the lens, camera sensor, and raw conversion or in camera software. However, MTFs multiply. To determine the MTF of the final output, one multiplies the MTF of the lens by that of sensor and software. This is discussed in simple terms on this site in section 3.2:

[a href=\"http://www.edmundoptics.com/techSupport/DisplayArticle.cfm?articleid=289]http://www.edmundoptics.com/techSupport/Di...m?articleid=289[/url]

Now even if your lens does not have the best MTF at a given frequency, a camera with a higher MTF will always give a better system MTF since this is the product of the lens MTF (which remains the same) and the camera MTF (which increases with the D20 because of its finer pixel pitch).

Norman Koren goes into more detail on his site

http://www.normankoren.com/Tutorials/MTF.html

The classical formula for system MTF is 1/r = 1/r1 + 1/r2 + ... + 1/rn, where r is the system MTF and r1, r2, etc are the MTFs of the lens, sensor and other components. This works only for low MTFs (around 10% such as used in USAF resolution charts). For higher frequencies, one must separate the frequency components with a Fourier transform, perform the multiplications, and then recombine the results with a cumbersome convolution process. This is beyond the capabilities of most of us, but the principle is understandable.

Another consequence of the multiplicative properties of MTFs is that the system MTF is lower than might be expected, since we are multiplying factors that are all less than unity. For example if the individual MTFs are 50% and there are three links in the chain, the overall MTF would be 12.5%

Bill
Logged
dmerger
Sr. Member
****
Offline Offline

Posts: 686


« Reply #4 on: July 27, 2006, 11:26:49 AM »
ReplyReply

Ray, what are the pixel dimensions of the photos from the 5D vs the 20D?
Logged

Dean Erger
Ray
Sr. Member
****
Offline Offline

Posts: 8908


« Reply #5 on: July 27, 2006, 11:36:48 AM »
ReplyReply

Quote
Ray, what are the pixel dimensions of the photos from the 5D vs the 20D?
[a href=\"index.php?act=findpost&pid=71862\"][{POST_SNAPBACK}][/a]

The 5D images are cropped to the same dimensions as the 20D images then interpolated to the same file size of 46.8MB. Essentially it's a comparison between 4.8m big pixels and 8m small pixels.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8908


« Reply #6 on: July 27, 2006, 12:10:04 PM »
ReplyReply

Quote
Another consequence of the multiplicative properties of MTFs is that the system MTF is lower than might be expected, since we are multiplying factors that are all less than unity. For example if the individual MTFs are 50% and there are three links in the chain, the overall MTF would be 12.5%
[a href=\"index.php?act=findpost&pid=71852\"][{POST_SNAPBACK}][/a]

Bill,
How does this explain the fact that detail (texture, high contrast fine B&W lines, low contrast broad lines etc) are hardly worse at f22? There should be a significant fall-off in MTF at all frequencies at f22 compared with f8, shouldn't there?

The following crops compare maximum line resolution for both cameras at f8 and f22. The number 100 on the chart is irrelevant. I didn't calculate a precise distance to the target that would enable me to derive an lp/mm rating, but if I were counting lines, I don't see the result would be much different at f22 or f8.


[attachment=846:attachment]                              [attachment=847:attachment]
Logged
bjanes
Sr. Member
****
Online Online

Posts: 2825



« Reply #7 on: July 27, 2006, 01:55:28 PM »
ReplyReply

Quote
Bill,
How does this explain the fact that detail (texture, high contrast fine B&W lines, low contrast broad lines etc) are hardly worse at f22? There should be a significant fall-off in MTF at all frequencies at f22 compared with f8, shouldn't there?

The following crops compare maximum line resolution for both cameras at f8 and f22. The number 100 on the chart is irrelevant. I didn't calculate a precise distance to the target that would enable me to derive an lp/mm rating, but if I were counting lines, I don't see the result would be much different at f22 or f8.
[attachment=846:attachment]                              [attachment=847:attachment]
[{POST_SNAPBACK}][/a]

Ray,

The 5d image is difficult to interpret because of aliasing and moire, so I will concentrate on the 20D image at f/8 and f/22. Actually, the MTF is much higher at f/8 as revealed by Norman Koren's quantitative methods using image J

[a href=\"http://www.normankoren.com/Tutorials/MTF5.html]http://www.normankoren.com/Tutorials/MTF5.html[/url]

Here is the 20D image at f/8 with readings taken from the top of the resolution chart. The maximum value is about 251 and the minimum value is 7. Calculating C(f) yields a value of 0.95 (C(f) = (Vmax-Vmin)/(Vmax+Vmin))

[attachment=848:attachment]

And here is the image at f/22. As you can see, the contrast is much lower, but the perceived resolution is about the same. The eye needs only a MTF of about 10% to resolve the lines:

[attachment=849:attachment]

C(f) is now 0.36.

To figure MTF I would need C0, and this is left to the reader.

Bill
Logged
dmerger
Sr. Member
****
Offline Offline

Posts: 686


« Reply #8 on: July 27, 2006, 02:32:23 PM »
ReplyReply

Ray, Iíve read more times than I can count about the superiority of the 20D pixel pitch, i.e. the 20D is capable of capturing more resolution than the 5D or 1Ds because of the 20Dís smaller pixel pitch.  Moreover, Iíve yet to see anyone dispute this proposition.  So, Iíll go out on a limb and dispute the rationale of this proposition.  (I donít own a DSLR so I canít speak from actual experience, but I donít believe that I need actual experience to poke holes in this proposition.)  

Ray, I postulate that your test only shows that, for a given sensor area and pixels of nearly equal quality, more pixels resolve more detail than fewer pixels.  I donít think your test shows that a smaller pixel pitch is superior.

Lets do a very simplified though experiment.  Assume we have a sensor size X with 100 pixels with a pixel pitch of Y.  We also have a sensor size ĹX with 100 pixels and a pixel pitch of ĹY.  Everything else is constant, i.e., the pixels and lenses over the entire sensor areas are of equal quality.  In this case, I postulate that pixel pitch Y will resolve the same detail as pixel pitch ĹY.  The smaller sensor needs the smaller pixel pitch to equal the resolution of the larger sensor because the image details on the smaller sensor are also smaller.  

In the real world, of course, pixels and lenses are rarely equal, especially when different sensor areas are involved.  Therefore, I wonít speculate whether in actual use a 20D is capable of resolving more detail than a 5D or 1Ds.  

Ray, I expect and hope that you and the other more technically inclined participants in this forum will let me know whether Iíve totally missed the boat on this one or if my reasoning is sound.
Logged

Dean Erger
spidermike
Full Member
***
Offline Offline

Posts: 117


« Reply #9 on: July 27, 2006, 04:16:52 PM »
ReplyReply

Quote
Ray,

In a test such as yours, the system MTF is being tested and this includes the lens, camera sensor, and raw conversion or in camera software. However, MTFs multiply. To determine the MTF of the final output, one multiplies the MTF of the lens by that of sensor and software. This is discussed in simple terms on this site in section 3.2:

http://www.edmundoptics.com/techSupport/Di...m?articleid=289

[a href=\"index.php?act=findpost&pid=71852\"][{POST_SNAPBACK}][/a]

I looked at the website quoted and....shuddered....

Quote
Every component within a system has an associated MTF and, as a result, contributes to the overall MTF of the system. This includes the imaging lens, sensor, capture boards, and cables, for instance.
(my emphasis)

Anyone who has even a passing interest in hi-fi will be fully aware of the claims, counterclaims, warfare and and total BS that surrounds so-called 'audiophile' cables.

Are we really heading that way....please God, No...
   
Logged
spidermike
Full Member
***
Offline Offline

Posts: 117


« Reply #10 on: July 27, 2006, 04:22:36 PM »
ReplyReply

Quote
The smaller sensor needs the smaller pixel pitch to equal the resolution of the larger sensor because the image details on the smaller sensor are also smaller.   

[a href=\"index.php?act=findpost&pid=71885\"][{POST_SNAPBACK}][/a]

Are they?
As has been recently postulated on a couple of threads (one of mine included), the image that hits the sensor is not smaller - the image projected by the lens is the same size in FF and cropped-frame cameras. It is just that the sensor, being physically smaller, covers a smaller part of the image projected by the lens.

So there is the theoretical possibility of greater resolution is the pixel pitch is smaller. I think the greater issue is that the smaller pixel pitch leads to greater electronic noise and so requires more in-camera processing.
Logged
dmerger
Sr. Member
****
Offline Offline

Posts: 686


« Reply #11 on: July 27, 2006, 06:04:01 PM »
ReplyReply

Spidermike, I donít see the logic in your analysis.  Using your logic, one could say that the image projected by a lens on a 35mm film camera is not smaller than the image projected on an 8x10 view camera.  You may be correct if you are prepared to only use a 35mm size crop from the 8x10, but isnít that approach a little bizarre?  

Perhaps I should have made it clear that I assumed that the full sensor area would be used in each case.
Logged

Dean Erger
Ray
Sr. Member
****
Offline Offline

Posts: 8908


« Reply #12 on: July 27, 2006, 07:55:57 PM »
ReplyReply

Quote
Ray, I postulate that your test only shows that, for a given sensor area and pixels of nearly equal quality, more pixels resolve more detail than fewer pixels.  I donít think your test shows that a smaller pixel pitch is superior.


Dmerger, I'm using the word superior here to describe an attribute which allows one camera (the 20D) to do something another camera (the 5D) cannot. There are other attributes of the 5D which could be described as superior to the 20D.

As a thought experiment, if you were stuck in a room with the 20D and 5D and had just one lens, which camera would allow you to capture the most information about the scene through the window?

The reason these tests are of interest to me is that ever since the introduction of the 1Ds we have had numerous claims that greater pixel density than that of the 1Ds will serve little purpose because lenses are simply not good enough.

In fact certain people have even implied that the performance of a lens at f22 is hardly better than that of a Coke bottle. Well, here's the proof of the pudding. Even at a Coke bottle aperture of f22, greater pixel density than that of the 1Ds and 5D can serve a purpose with ordinary, current 35mm lenses.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8908


« Reply #13 on: July 27, 2006, 09:14:15 PM »
ReplyReply

Quote
And here is the image at f/22. As you can see, the contrast is much lower, but the perceived resolution is about the same. The eye needs only a MTF of about 10% to resolve the lines:
[a href=\"index.php?act=findpost&pid=71883\"][{POST_SNAPBACK}][/a]


Bill,
I think that is the explanation. We're now back to a previous discussion about the extent to which the lower over all MTF at f22 can be augmented with contrast enhancement and sharpening techniques.

Both images below, comparing f8 with f22 on the 20D, have been sharpened, but the f22 image sharpened more and local contrast enhancement applied. I've overdone the f22 contrast a bit, but the images are surprisingly close, to my eyes, considering these are 300% enlargements.

There's no doubt though that the f8 image is slightly better, cleaner. The f22 image shows a break-up of the finer lines sooner.

BTW, the irregularity of the contrast and width of these lines is due to imperfect printing, but I'm not sure if this is caused by inappropriate gamma, dot gain or simply that I didn't use the 2880 dpi setting. In these examples, it doesn't really matter because the irregularity is a detail feature in itself and shows up at both f8 and f22.

[attachment=850:attachment]
Logged
dmerger
Sr. Member
****
Offline Offline

Posts: 686


« Reply #14 on: July 27, 2006, 09:17:58 PM »
ReplyReply

Ray, it depends.  If you are at the window with an unobstructed view, and ďmost informationĒ means capturing as much of the exterior scene as possible, then the 5D wins hands down.  If you are far from the window and a large part of the 5D frame is of the interior, then the 20D may capture a little more resolution of the exterior.  In the first example, however, the 20D captures ZERO information outside its field of view.  In the second example, the 5D captures the same field of view of the exterior but with a little less resolution.  So, without more information, Iíd say that the 5D would capture the most information, no contest.
Logged

Dean Erger
Ray
Sr. Member
****
Offline Offline

Posts: 8908


« Reply #15 on: July 27, 2006, 09:25:59 PM »
ReplyReply

Quote
Ray, it depends.
[a href=\"index.php?act=findpost&pid=71931\"][{POST_SNAPBACK}][/a]

You're stuck, whether close to the window or not. You need to capture as much information as possible about enemy troop positions in the distance. You can take as many shots as your flash card allows. You have one lens only. Which camera do you use?
Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #16 on: July 27, 2006, 09:38:42 PM »
ReplyReply

Quote
You're stuck, whether close to the window or not. You need to capture as much information as possible about enemy troop positions in the distance. You can take as many shots as your flash card allows. You have one lens only. Which camera do you use?
[a href=\"index.php?act=findpost&pid=71932\"][{POST_SNAPBACK}][/a]

The way people respond to these types of comparisons really shows what kind of photography they do.  Anyone who has ever done much wildlife photography knows that the longest telephotos become feeble when the subject is either small or distant, and people with this kind of experience know what its like to have a large frame of nothing.
« Last Edit: July 27, 2006, 09:56:51 PM by John Sheehy » Logged
dmerger
Sr. Member
****
Offline Offline

Posts: 686


« Reply #17 on: July 27, 2006, 09:48:56 PM »
ReplyReply

Neither, I run like hell!

Ray, I see your point, but I don't see what it proves.  We can create a lot of scenarios where a 5D would be better and others where a 20D would be better, e.g. whether a wide field of view is desired or you donít have a long enough lens and want a close-up. In your scenario, the 20D would be best, but I donít think it has any bearing on the ďsuperiority of the 20D pixel pitchĒ.
Logged

Dean Erger
Ray
Sr. Member
****
Offline Offline

Posts: 8908


« Reply #18 on: July 28, 2006, 12:30:30 AM »
ReplyReply

Quote
Neither, I run like hell!

Ray, I see your point, but I don't see what it proves.  We can create a lot of scenarios where a 5D would be better and others where a 20D would be better, e.g. whether a wide field of view is desired or you donít have a long enough lens and want a close-up. In your scenario, the 20D would be best, but I donít think it has any bearing on the ďsuperiority of the 20D pixel pitchĒ.
[a href=\"index.php?act=findpost&pid=71942\"][{POST_SNAPBACK}][/a]
I

It has no bearing on the superiority of one camera in ralation to another in general terms, but it does have a bearing in specific terms. As the saying goes, best tool for the job.
Logged
skid00skid00
Jr. Member
**
Offline Offline

Posts: 53


« Reply #19 on: July 29, 2006, 04:46:54 PM »
ReplyReply

Don't forget that the 5D and 20D have much stronger AA filters.

1Ds produces VERY sharp single-pixels.
Logged
Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad