Ad
Ad
Ad
Pages: [1] 2 3 ... 7 »   Bottom of Page
Print
Author Topic: 1Ds2 versus 5D  (Read 57811 times)
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« on: July 01, 2006, 08:12:50 AM »
ReplyReply

I've read many threads on the net discussing the resolution differences between the 1Ds2 and the 5D and they've all seemed very inconclusive, some photographers claiming the 5D is actually superior resolution-wise and some claiming the 1Ds2 has the edge if the print is large enough.

Having recently received Michael's 'Measuring Megabytes' disc, I decided to have a close look at the 1Ds2 and 5D files. I upressed the 5D file first, using bicubic, so I was comparing same size files and was surprised that I couldn't find one iota of detail, one smudge however faint, that was resolved on the 1Ds2 image but not on the 5D image. This is remarkable, but can't be right. So I kept enlarging the image on screen to a point where the equivalent print size of the whole image would be ridiculously large, far bigger than any printer I know of could handle.

And guess what! At 400% enlargement I found some micro detail on the dollar bill that was better defind on the 1Ds2 image. Here's the comparison below. No sharpening applied to either image of course.

[attachment=777:attachment]

Wow!  
Logged
Eric Myrvaagnes
Sr. Member
****
Offline Offline

Posts: 8217



WWW
« Reply #1 on: July 01, 2006, 09:36:38 AM »
ReplyReply

Yikes, Ray!

Thanks for pointing that out. I guess I'll keep saving my pennies for a 5D -- unless, of course, I decide to go into the billboard-size counterfeiting business.    

Eric
Logged

-Eric Myrvaagnes

http://myrvaagnes.com  Visit my website. New images each season.
Tim Gray
Sr. Member
****
Offline Offline

Posts: 2002



WWW
« Reply #2 on: July 01, 2006, 10:23:17 AM »
ReplyReply

What's interesting is to take that same crop from the P40, P25, 1ds2 and 5d and see the improvement in microdetail as you move up the food chain.

Dare I define microdetail as "detail that becomes apparent only at outrageously large print sizes"  

I've printed out these 4 captures (f11's) at my measely 17"x24" and even with a magnifying glass I'm about 75% sure the difference I see are entirely in my head.

I usually roll my eyes a bit at the "how large can I print" posts, but find myself somehow drawn to the opposite question - "how large do I have to print before the difference becomes worthwhile?"
Logged
John Sheehy
Sr. Member
****
Offline Offline

Posts: 838


« Reply #3 on: July 01, 2006, 01:08:05 PM »
ReplyReply

Quote
And guess what! At 400% enlargement I found some micro detail on the dollar bill that was better defind on the 1Ds2 image. Here's the comparison below. No sharpening applied to either image of course.
Wow! 
[a href=\"index.php?act=findpost&pid=69581\"][{POST_SNAPBACK}][/a]

Actually, that looks like a pretty big difference to me, considering the 1DsmkII is only expected to have 14% more linear resolution.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #4 on: July 01, 2006, 08:59:09 PM »
ReplyReply

Quote
Actually, that looks like a pretty big difference to me, considering the 1DsmkII is only expected to have 14% more linear resolution.
[a href=\"index.php?act=findpost&pid=69594\"][{POST_SNAPBACK}][/a]

It is a big difference, but it's a big difference on a huge print that would be far larger than any giclee type printer could handle.

On my 19" monitor with screen resolution of 1280x1024, a 400% enlargement of the entire image equates to a print size of approximately 17.5 feet wide. If I view such a print from a distance equal to the print's diagonal, then such differences that I see viewing the 400% crop from a distance equal to my monitor's diagonal, disappear completely. In fact, I only have to step back about 12ft from my monitor for those differences to become undiscernible.

Now I know people have a habit of walking up close to a print, whatever the size. I some times do this with billboards to check out the size of the ink dots (just out of curiosity   ), but I submit, if these two 17.5ft wide prints were side by side on a gallery wall and you wanted to compare micro detail, by the time you'd walked over to the other print, you'd probably have forgotten the precise nature of the micro detail you'd been examining just 30 seconds ago. (Maybe not   ).

I haven't made any prints of these crops yet, as Tim suggests, but at 67% enlargement on my monitor, the full size image is as big as my Epson 7600 can handle without borders on all 4 sides (ie, 36x24").

At this degree of enlargement (67%) the differences clearly visible at 400% are no longer discernible. It's my experience that what I can't see on the monitor, I can't see on the print (excluding subtle hues of color that might be outside the gamut of the monitor but inside the gamut of the ink and paper).

By the way, that 14% increase in resolution that the 1Ds2 is supposed to have is a theoretical calculation that can only be realised with a perfect lens. It represents an increase in sensor resolution as opposed to system resolution. In practice, the resolution increase will always be smaller and herein lies the difficulty as I see it.

The 30% increase in pixel count, of the 1Ds2, appears fairly substantial and creates the impression that the resolution increase should be substantial. However, the resolution increase in terms of lp/mm along one dimension is a theoretical maximum of 14% and probably no more than half that in practice.

In my view, some people incorrectly interpret this small increase (say 7% in practice) as an indication that current lenses are simply not good enough to justify any further increase in pixel density. My interpretation is, a 30% increase in pixel count is simply too trivial to make a worthwhile difference. You need to double pixel count to get an improvement that hits you in the eye. Such improvement, however, would be overkill for the printing of uncropped images at normal or average print sizes, but great for those wanting to get the most from their longest lenses, or for those occasions when an interesting composition can be found by heavily cropping an image.
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2866



« Reply #5 on: July 01, 2006, 09:25:46 PM »
ReplyReply

Quote
I've read many threads on the net discussing the resolution differences between the 1Ds2 and the 5D and they've all seemed very inconclusive, some photographers claiming the 5D is actually superior resolution-wise and some claiming the 1Ds2 has the edge if the print is large enough.

Having recently received Michael's 'Measuring Megabytes' disc, I decided to have a close look at the 1Ds2 and 5D files. I upressed the 5D file first, using bicubic, so I was comparing same size files and was surprised that I couldn't find one iota of detail, one smudge however faint, that was resolved on the 1Ds2 image but not on the 5D image. This is remarkable, but can't be right. So I kept enlarging the image on screen to a point where the equivalent print size of the whole image would be ridiculously large, far bigger than any printer I know of could handle.

And guess what! At 400% enlargement I found some micro detail on the dollar bill that was better defind on the 1Ds2 image. Here's the comparison below. No sharpening applied to either image of course.

Wow! 
[a href=\"index.php?act=findpost&pid=69581\"][{POST_SNAPBACK}][/a]

I do not think that the comparison is valid. When an image is resampled, there is always some loss of sharpness. That's why sharpening is usually recommended after resampling. The resampling of the 5D image undoubtedly softened it. If you look carefully at the arc of the lines on the EOS 1Ds image near the arrowhead, you see that the pixels of the engraving line sweep out a smooth arc, whereas in the 5D image they are irregularly spaced above and below the arc. This indicates interpolation error.

Secondly, a great deal of alaising is present in both images, and alaising can add false detail or obscure real detail depending on the frequency mismatch between the image and the sensor. Take a look at the resolution charts on DPReview; near Nyquist the lines of the test chart are very poorly defined and false lines come and go.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #6 on: July 01, 2006, 11:28:21 PM »
ReplyReply

Quote
I do not think that the comparison is valid.
[a href=\"index.php?act=findpost&pid=69615\"][{POST_SNAPBACK}][/a]

I do not see what else I can do to make it more valid. If I sharpened the 5D image after upsampling, I think one would have even more reason for claiming the comparison is invalid. The fact is, whether I upsample with bicubic, bicubic smoother, bicubic sharper or even Genuine Fractals, I cannot get that extra detail visible in the 1Ds2 shot at 400%. Nor can I with any sharpening routine that I've tried.

I have no reason to suppose the extra detail in the 1Ds2 image is due to aliasing artifacts, and any aliasing artifacts in the 5D image, I would suggest, would not be visible on a 36x24" print.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #7 on: July 02, 2006, 12:51:17 AM »
ReplyReply

Here's the same size crop of the 400% enlarged image using 'bicubic sharper' to upsample the 5D image, plus a minimal amount of sharpening with Focus Magic. The 1Ds2 image remains unsharpened.

So you can get an idea of the extent of aliasing artifacts in both images, I've included a crop of the Betterlight image, which of course is very, very much better. I wonder if we'll ever have such quality in a full frame 35mm DSLR, whilst I'm still alive.   .

You can see I haven't bothered doing any color corrections in the 35mm images. The Betterlight image was already converted on the disc.

[attachment=781:attachment]

Well, I suppose I'd better make an attempt at a better white balance, otherwise my methodology might get criticised   . Is this better?

[attachment=782:attachment]
« Last Edit: July 02, 2006, 01:47:03 AM by Ray » Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2866



« Reply #8 on: July 02, 2006, 07:52:10 AM »
ReplyReply

Quote
I do not see what else I can do to make it more valid. If I sharpened the 5D image after upsampling, I think one would have even more reason for claiming the comparison is invalid. The fact is, whether I upsample with bicubic, bicubic smoother, bicubic sharper or even Genuine Fractals, I cannot get that extra detail visible in the 1Ds2 shot at 400%. Nor can I with any sharpening routine that I've tried.

I have no reason to suppose the extra detail in the 1Ds2 image is due to aliasing artifacts, and any aliasing artifacts in the 5D image, I would suggest, would not be visible on a 36x24" print.
[a href=\"index.php?act=findpost&pid=69623\"][{POST_SNAPBACK}][/a]

If I were doing your test, I would downsample the 1Ds2 MII image using bicubic and upsample the 5D image also using bicubic to a common file size. In that way one camera would not have any advantage. Adobe recommends bicubic smoother for upsampling and bicubic sharper for downsampling, and you might want to try that too, but it would introduce another variable.

The 36 X 24 inch print would be a more reasonable print size, but weren't you talking about a 72 foot image for your onscreen views?

If resolution is your only concern, it is not surprising that the the higher resolution camera would come out better. However, perceived image quality does not relate well to resolution at very high frequency with a very low MTF, but rather to the resolution at a MTF of 50%. In the 21st century one should include the MTF for any resolution figure. Your greatly enlarged images are showing small details at a MTF probably around 10%, which is not considered to contribute much to perceived image quality.

I downloaded the resolution charts from DPReview for the 5D and EOS 1Ds and determined the resolution at MTF 50 with Imitest. The results are as follows:

5D: vertical 2192 lw/ph, horizontal 2181 lw/ph
1Ds2: vertical 2136 lw/ph, horizontal 2234 lw/ph

lw/ph = line pairs / picutre height

The design advances of the 5D appear to compensate for somewhat lower pixel density.
« Last Edit: July 07, 2006, 07:41:23 AM by bjanes » Logged
Jack Flesher
Sr. Member
****
Offline Offline

Posts: 2595



WWW
« Reply #9 on: July 02, 2006, 08:46:37 AM »
ReplyReply

Quote
I've included a crop of the Betterlight image, which of course is very, very much better.
[attachment=781:attachment]

Interesting find Ray!

(And now you also know why I recently adopted a Betterlight back  )
Logged

Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #10 on: July 02, 2006, 10:05:51 AM »
ReplyReply

Quote
If I were doing your test, I would downsample the 1Ds2 MII image using bicubic and upsample the 5D image also using bicubic to a common file size. In that way one camera would not have any advantage.


And if you did that, I would criticise you mercilessly.  

The purpose of this comparison was to find out if there are any circumstances in which the 1Ds2 might show some resolution advantage, however small. My purpose was not to throw away some 1Ds2 image detail in order to equalise both images. In any case, all prints larger than 14x21" would require upsampling of both images at the recommended 240ppi resolution.

Quote
The 36 X 24 inch print would be a more reasonable print size, but weren't you talking about a 72 foot image for your onscreen views?


You're just a little bit out there, Bill. I was talking about a 17.5ft image, but I imagine those subtle differences would be visible in a much smaller print, say 8ft wide, but not 3 ft wide.

Quote
Your greatly enlarged images are showing small details at a MTF probably around 10%, which is not considered to contribute much to perceived image quality.


I know. You're not getting the impression I'm suggesting rushing out to buy a 1Ds2 on the basis it can produce such a miniscule improvement over the 5D, are you?  
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #11 on: July 02, 2006, 10:54:52 PM »
ReplyReply

Quote
(And now you also know why I recently adopted a Betterlight back  )
[a href=\"index.php?act=findpost&pid=69635\"][{POST_SNAPBACK}][/a]

Jack,
I'm a firm believer in using the best tool for the job. I guess you will now be able to produce stunningly detailed shots of the Grand Canyon at the largest size a giclee printer can manage. What is that size? 4ft high, 6 ft high, 8ft high? Do you have such a printer?
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2866



« Reply #12 on: July 03, 2006, 09:16:22 AM »
ReplyReply

Quote
And if you did that, I would criticise you mercilessly.   

The purpose of this comparison was to find out if there are any circumstances in which the 1Ds2 might show some resolution advantage, however small. My purpose was not to throw away some 1Ds2 image detail in order to equalise both images. In any case, all prints larger than 14x21" would require upsampling of both images at the recommended 240ppi resolution.
[a href=\"index.php?act=findpost&pid=69636\"][{POST_SNAPBACK}][/a]

Ray, you don't want to throw away 1Ds2 detail, but have no reservations about throwing away 5D detail? I don't have the LL DVD and do not know in what format the images are supplied, but I inferred they were raw with as trial version of Capture 1. If this is the case, how did you process the images?

I read Bill Atkinson's test methodology with some interest since he is a scientist and software expert as well as photographer. In his testing, he used Capture 1 at default settings with no sharpening or noise reduction. He cropped a selected area of the images to 8000 x 6000 pixels at 360 dpi with bicubic interpolation, normalized the images for contrast and color using the color checker squares, and then applied a standardized sharpening, which was necessary to bring out the differences in the captures.

Your test would be more valid if you followed Bill's methodology and upresed  both images to a constant large size before making the comparison. You might also check the contrast. The 1Ds2 appears to have more contrast. Finally, some sharpening is necessary in a realistic test, since it would always be applied to a real print.

Quote
I know. You're not getting the impression I'm suggesting rushing out to buy a 1Ds2 on the basis it can produce such a miniscule improvement over the 5D, are you? 
[a href=\"index.php?act=findpost&pid=69636\"][{POST_SNAPBACK}][/a]

No, I do not get that impression. The 5D's larger pixel size give it an advantage in dynamic range and lower noise, which is an advanatage in many situations. It is interesting that in the 30D Canon did not match the resolution of the D200, but opted to keep the larger pixel size.
Logged
Jack Flesher
Sr. Member
****
Offline Offline

Posts: 2595



WWW
« Reply #13 on: July 03, 2006, 09:30:22 AM »
ReplyReply

Quote
but I imagine those subtle differences would be visible in a much smaller print, say 8ft wide, but not 3 ft wide.
[a href=\"index.php?act=findpost&pid=69636\"][{POST_SNAPBACK}][/a]

Actually, those minor differences are visible in prints 24" wide.  

But this is where the "which camera is better" debate always falls apart.  If you want to see those differences, you have to compare large prints (and by large, I mean over 16" on the short dimension).  This requires you optimise each camera's file to the final output resolution. This in turn requires in most cases completely different settings from raw conversion through final print sharpening.  

IMO the only real way to settle it is to get a 5D shooter and a 1Ds2 shooter together, each takes the same image with the same lens on their camera, then each process their own file as they deem optimal for their camera, then both files are printed out on the same printer and compared.  

Of course I did that for myself when I first got the 5D and I saw distinct advantages to the 1Ds2 file.  However I had an early 5D and it is certainly possible that 1) I had a bad copy and/or 2) due to my more limited experience with the 5D I did not process its file as optimally as the 1Ds2's.   But I followed it up with a friend and his 5D a few months later and we saw the same result -- FWIW he sold his 5D and is now shooting the 1Ds2.  

Admittedly, the differences while visible are small and may not be significant enough for others to even worry about -- a decision wholly up to the artist IMO.
« Last Edit: July 03, 2006, 09:42:02 AM by Jack Flesher » Logged

Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #14 on: July 03, 2006, 05:51:29 PM »
ReplyReply

Quote
If this is the case, how did you process the images?


I haven't installed the trial version of Capture One that's included on the disc. This is an initial look at just a few of the test shots addressing an issue that's been puzzling me for a while. Jack Flesher will understand. This issue was hotly debated on the old Galbraith forum.

I used ACR 3.4 to convert both RAW files, as shot with no adjustments, WB correction or sharpening. In my experience, sharpening does not create detail where none exists. It merely enhances detail that is already there. As stated before, my purpose was to look for any additional detail in the 1Ds2 file and assess its significance in a real world image (if one can really call it that) as opposed to a test chart of high contrast lines.

Now I could try lots of different types of processing, comparing C1 with ACR with RSP, upsampling both images to various but equal resolutions, using different types of interpolation algorithms etc etc, but I've got my own tests to do which are more relevant to my own purposes. However, if anyone else has got the time to experiment with these 2 images in different ways which can somehow change the differences between them in some significant way, I'd be interested in seeing the result.

Quote
Your test would be more valid if you followed Bill's methodology and upresed  both images to a constant large size before making the comparison. You might also check the contrast. The 1Ds2 appears to have more contrast.


I can't see the logic of this since I was comparing just 2 images. I threw in the Betterlight image so you might get a clearer idea of the artifacts in the 5D and 1Ds2 image. Sharpening doesn't create additional detail. Nor does interpolation.

You seem to be getting hung up on the cosmetics of this, Bill.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #15 on: July 03, 2006, 06:01:36 PM »
ReplyReply

Quote
Actually, those minor differences are visible in prints 24" wide. 


With or without loupe, Jack?  

Quote
IMO the only real way to settle it is to get a 5D shooter and a 1Ds2 shooter together, each takes the same image with the same lens on their camera, then each process their own file as they deem optimal for their camera, then both files are printed out on the same printer and compared. 


If the 5D shooter was me and the 1Ds2 shooter you, then I'm sure the 1Ds2 print would look much better, Jack.  
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2866



« Reply #16 on: July 03, 2006, 06:30:41 PM »
ReplyReply

Quote
I can't see the logic of this since I was comparing just 2 images. I threw in the Betterlight image so you might get a clearer idea of the artifacts in the 5D and 1Ds2 image. Sharpening doesn't create additional detail. Nor does interpolation.

You seem to be getting hung up on the cosmetics of this, Bill.
[a href=\"index.php?act=findpost&pid=69707\"][{POST_SNAPBACK}][/a]

IMHO, this is not cosmetics, but a flaw in your test procedure. Interpolation can and usually does degrade image quality and was done on one Canon image but not the other, so the comparison is invalid. Sharpening does not add any detail, but it does bring out what detail is present and was considered desirable by Bill Atkinson for a comparison of the images. Why don't you rebut his suggtions also?
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8939


« Reply #17 on: July 03, 2006, 08:34:14 PM »
ReplyReply

Geez!, Bill, you're an exasperating guy. You know that, don't you.  

Quote
Interpolation can and usually does degrade image quality


There's no noticeable image degradation when upsampling the 72MB 5D image to 95MB. I know because I've got the images and I'm looking at them. When comparing 2 or more images of the same FoV but different pixel densities, it is correct methodology to upsample the images with the smaller native resolution to that of the largest image. This is what Bill Atkinson has done and it's what I've done. We've both upsampled the image(s) to the size of the largest, highest resolution image which we are comparing, which in my comparison is the 1Ds2 and in Bill's comparison is the 6000x8000 Betterlight.

In general, interpolation gives an impression of additional softness because the interpolated image is larger. The larger any image is interpolated, the softer it will appear at that larger size and conversely the smaller any image, the sharper it will appear to the point where one has reached the limit of one's eyesight. For example, if I compare the 72MB 5D image with itself when interpolated to 95.1MB, at equal magnifications on screen the interpolated image will be larger and appear very marginally softer. If I reduce the degree of enlargement of the interpolated image so the 2 crops are of identical size, which in this case would be 400% for the uninterpolated image and something like 358% for the interpolated image, there is no discernible difference between the 2 images, no noticeable degradation. This however, might not be the case when images are interpolated by a huge degree.

Quote
Sharpening does not add any detail, but it does bring out what detail is present


No amount of sharpening of the 5D image can restore that small amount of extra detail present in the 1Ds2 image. I've tried it. I've also provided you with 2 sets of comparisons, one with no sharpening at all to either image, which makes the 1Ds2 image appear marginally sharper and more contrasty, and the other with sharpening applied to the 5D image, which makes the 5D image appear marginally sharper and more contrasty.

These results seem pretty conclusive to me and I see no flaw in the methodology in relation to the purpose of the comparison. If there is a flaw, then someone else will have to demonstrate the flaw by producing a different result. I'm sorry I can't email you the RAW images. I'm on a dial-up 56k connection. If you want to pursue this matter, I suggest you buy the disc.
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2866



« Reply #18 on: July 03, 2006, 10:10:58 PM »
ReplyReply

Quote
Geez!, Bill, you're an exasperating guy. You know that, don't you.   
[a href=\"index.php?act=findpost&pid=69714\"][{POST_SNAPBACK}][/a]

Ray,

Yes, others have made that point also  .  I could say the same about you, but I always learn somethings from our discussions and sometimes play the devil's advocate when something in the discussion does not seem quite right to me. However, at this point can think of no reasonable further objections, and I accept that you have made your case. Thanks for your patience, sound reasoning, and civility.
Logged
Jack Flesher
Sr. Member
****
Offline Offline

Posts: 2595



WWW
« Reply #19 on: July 04, 2006, 09:04:55 AM »
ReplyReply

Quote
Jack Flesher will understand.

SNIP

I used ACR 3.4 to convert both RAW files, as shot with no adjustments, WB correction or sharpening.

Jack Flesher does understand.  

But he also knows that ACR cannot extract the maximum detail from a 1Ds2 file.  Love the color, love the workflow, but the detail simply aint there.  However, it seems it CAN extract all the detail present on a 5D file -- or at least Pixmantec's Rawshooter does not glean much more from a 5D file than ACR does.  But on a 1Ds2 file, there is no contest, the 1Ds2 files come alive.  Download the freebie Rawshooter if you don't believe me.  

So I also understand Adobe's recent aquisition of the Pixmantec technology.
« Last Edit: July 04, 2006, 09:06:56 AM by Jack Flesher » Logged

Pages: [1] 2 3 ... 7 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad