Ad
Ad
Ad
Pages: « 1 ... 6 7 [8]   Bottom of Page
Print
Author Topic: If its not megapixels what is it?  (Read 20209 times)
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3585


« Reply #140 on: May 23, 2011, 07:15:58 AM »
ReplyReply

My reference to printing at 360 dpi came from tests with my own printer and it's driver (An Epson 3800 set to Superfine photo 2880 x 1440 dpi). A test pattern of vertical lines; horizontal lines and a checkerboard - each 1 pixel wide is reproduced perfectly at 360 ppi but suffers from artifacts at 359ppi or less or at 361ppi or more. (Sending 720 ppi does not reproduce correctly either).

Hi Dave,

That sounds suspect. Could it be that resampling was used to reach the 359 / 361 PPI? If so how was the resampling done?

The same goes for the 720 PPI, which is the native resolution of your printer (when using glossy paper and the appropriate driver settings). Others report superior resolution at 720 PPI vs 360 PPI.

Cheers,
Bart
Logged
DaveRichardson
Newbie
*
Offline Offline

Posts: 14


« Reply #141 on: May 23, 2011, 07:48:49 AM »
ReplyReply


How did you conclude that 360dpi prints was reproduced perfect? Using your eyesight closeup, a scanner, a camera using macro lense? Is it safe to conclude that the clean, apparently unaliased image that resulted was the result of an end-to-end 1:1 pipeline, or could it be that you were actually viewing an aliased image that happened to give you clean patterns similar to the input (at e.g. an integer multiplum lower frequency)?


Thanks for the response.

My use of the word "perfect" is incorrect. With my eyes, wearing reading glasses, I could percieve a pattern of lines and dots at 360 dpi. It took a magnifying glass to show that this was a reproduction of the 1 pixel patterns I had sent and did not appear to show any particular artifacts.

For the 359 and 361 ppi tests - the artifacts were visible with the naked eye at 8-10 feet - they appeared as a repeating pattern of dark / light lines. It would seem that the printer driver is doing some resampling.


For this test, it was possible to generate content tailormade for the printer input resolution, where pixel-to-pixel difference is 100%. For realistic scenarios, one will grab a e.g. 20 megapixel image from camera, and decide on some physical print size (e.g. A3). Then content would have to be resampled at least once, meaning that bar-code patterns will loose some acuity.



I agree - to get through the end to end chain of camera to my printer, with no resampling (which with the exception of some specific integer sizes, will always introduce some impact on the image, and for an A2 print would require a 50MP sensor that samples each colour at each pixel site. My original point was that even the 80M sensors that exist today are not capable of doing that due to the use of the bayer pattern.



I agree that your test seem to indicate that there is that for your setup some gain in choosing 360dpi as output format. Whether your images can use that gain, and whether any given human can perceive it at any given print size/distance is open for debate.

-h

In the real world, it is rare that we do produce images that are not scaled (or have not used specific integers for up-scaling). However the thinking behind my original comment was that there is printable data from the 80Mp sensors that is not being captured in smaller sensors, and even with the 80Mp sensor we have not yet reached the point where we are capturing more data than is printable at A2, even on my old printer. I totally agree that it remains debatable what can be seen in printed images from normal viewing distances - but not having had the good fortune to see images from these sensors directly I was only able to hypothesise on why those who have used the sensors might be seeing a real difference.

Dave
Logged
DaveRichardson
Newbie
*
Offline Offline

Posts: 14


« Reply #142 on: May 23, 2011, 08:09:31 AM »
ReplyReply



That sounds suspect. Could it be that resampling was used to reach the 359 / 361 PPI? If so how was the resampling done?


Cheers,
Bart

Hi Bart
I was careful not to resample when I resized the image from 360ppi to 359/358/361/362 and 720 ppi for testing. This was confirmed with no impact on screen only the reported image dimensions changing. Then I just sent each image to the printer driver.
It looks like my driver (Epson) is doing some internal resampling at resolutions other than 360 dpi

I just repeated the 720 dpi test to ensure I had not made a mistake. I was using Premium Glossy paper and had the printer set to 2880 x 1440dpi (Superfine). The  printer did not reproduce the fine horizontal and vertical lines that were seen when sending 360 dpi. Just solid areas of white and black. It may be that another setting in the driver may be required for 720 dpi.

Dave
Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1677


« Reply #143 on: May 23, 2011, 08:28:34 AM »
ReplyReply

I agree - to get through the end to end chain of camera to my printer, with no resampling (which with the exception of some specific integer sizes, will always introduce some impact on the image, and for an A2 print would require a 50MP sensor that samples each colour at each pixel site. My original point was that even the 80M sensors that exist today are not capable of doing that due to the use of the bayer pattern.

In the real world, it is rare that we do produce images that are not scaled (or have not used specific integers for up-scaling). However the thinking behind my original comment was that there is printable data from the 80Mp sensors that is not being captured in smaller sensors, and even with the 80Mp sensor we have not yet reached the point where we are capturing more data than is printable at A2, even on my old printer. I totally agree that it remains debatable what can be seen in printed images from normal viewing distances - but not having had the good fortune to see images from these sensors directly I was only able to hypothesise on why those who have used the sensors might be seeing a real difference.

Dave
As long as one is allowing any print size, there will always be a threshold where at one side the printer is "out-resolving" the camera, and at the other side the camera is outresolving the printer. The larger your prints, the greater the chance that your printer is capable of reproducing details that your camera cannot capture.

I believe that natural images and image processing pipelines tend to behave in such a way that classical sampling theory/Nyquist is applicable (unless excessive sharpening is applied). I believe that synthetic test-patterns like yours belong to a different cathegory, and great care should be taken when comparing response to such test-patterns vs natural images (perhaps band-limited swept sines would be a better choice).

I believe that it is fair to assume that people tend to view A3 sized prints at a greater distance than A4 sized prints. This tendency may not be enough to compensate for the change in print size (i.e. lp/ph may be more important in large images than in small ones).

The human visual system is often said to be limited to 1 minute of arc (angular subtended), and we have a limited ability to focus very near to some object. I believe that this can be used to set a hard limit on how much real resolution is of use for 20/20 vision people not using magnifyers.

-h
« Last Edit: May 23, 2011, 08:31:51 AM by hjulenissen » Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3585


« Reply #144 on: May 23, 2011, 09:40:29 AM »
ReplyReply

Hi Bart
I was careful not to resample when I resized the image from 360ppi to 359/358/361/362 and 720 ppi for testing. This was confirmed with no impact on screen only the reported image dimensions changing. Then I just sent each image to the printer driver.
It looks like my driver (Epson) is doing some internal resampling at resolutions other than 360 dpi

Hi Dave,

It's still very strange and contrary to what others report. Do you by any chance have borderless printing activated (that will cause resampling by the printer driver, but not very high quality resampling)? I assume you print from Photoshop?

Here is a test file that can be used to find out if the 720 PPI is resolved by the printer (which it normally should if the printheads are aligned).

Cheers,
Bart
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3585


« Reply #145 on: May 23, 2011, 09:47:16 AM »
ReplyReply

For the 359 and 361 ppi tests - the artifacts were visible with the naked eye at 8-10 feet - they appeared as a repeating pattern of dark / light lines. It would seem that the printer driver is doing some resampling.

Hi Dave,

This is a clear indication of aliasing artifacts, most likely caused by resampling. Make sure that borderless printing is not active.

Cheers,
Bart
Logged
DaveRichardson
Newbie
*
Offline Offline

Posts: 14


« Reply #146 on: May 23, 2011, 10:52:11 AM »
ReplyReply

Hi Bart
Thanks for the test print links - I'll try them later and come back.
In the meantime I can confirm I did not have borderless printing enabled. The 359; 360 ; 361 ppi prints gave a result as I expected as I had been told previously that the EPson driver tried to scale if sent a resolution other than 360/720ppi. It was the 720 ppi print that surprised me.

Dave
Logged
DaveRichardson
Newbie
*
Offline Offline

Posts: 14


« Reply #147 on: May 23, 2011, 12:57:14 PM »
ReplyReply

Hi Bart
I tried the test images from your link and found the following :
To get the driver to accept 720 ppi I needed to set the "Finest Detail" setting in the driver. It would then accept and correctly print the 720 ppi tests. I had not checked this previously as Epson recommended it for graphics only.

Eric Chan's website describes the Epson driver and describes that with "Finest detail" unchecked everything sent to the driver is resampled to 360 ppi and with it checked, it is resampled to 720 dpi. This explains my findings at 359ppi and 361 dpi.

So is the difference between 360 ppi and 720 PPI visible ? 
In test prints with 1 pixel wide lines and more especially lines that are just off vertical then a difference between 720ppi and 360 ppi is clearly visible to the naked eye (or in my case with my glasses on). However in real photographs I struggled to see a difference between 360 ppi and 720 ppi even with a magnifying glass. I will therefore leave it unchecked and continue to send 360 dpi to this printer.

To work toward an A2 print end to end at 720 ppi would require image capture at 200Mp (preferably with 3 colours directly captured at each pixel) - a year or two away  I think   Smiley

Dave
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3585


« Reply #148 on: May 23, 2011, 01:20:19 PM »
ReplyReply

Hi Bart
I tried the test images from your link and found the following :
To get the driver to accept 720 ppi I needed to set the "Finest Detail" setting in the driver. It would then accept and correctly print the 720 ppi tests. I had not checked this previously as Epson recommended it for graphics only.

Eric Chan's website describes the Epson driver and describes that with "Finest detail" unchecked everything sent to the driver is resampled to 360 ppi and with it checked, it is resampled to 720 dpi. This explains my findings at 359ppi and 361 dpi.

Hi Dave,

Glad you've been able to solve that one. I assume the aliasing artifacts at lower PPIs are also reduced.

Quote
So is the difference between 360 ppi and 720 PPI visible ? 
In test prints with 1 pixel wide lines and more especially lines that are just off vertical then a difference between 720ppi and 360 ppi is clearly visible to the naked eye (or in my case with my glasses on). However in real photographs I struggled to see a difference between 360 ppi and 720 ppi even with a magnifying glass.

Just remember that resampling to 720 PPI also allows to do the print sharpening at that native resolution. It will keep gradients smoother, and you can increase the edge acutance even further than at 360 PPI. Of course whether it matters a lot is subject dependent and it also depends on how much real data there is to begin with.

Quote
To work toward an A2 print end to end at 720 ppi would require image capture at 200Mp (preferably with 3 colours directly captured at each pixel) - a year or two away  I think   Smiley

It's only a matter of time, but in the mean time a good image upsampling program can help a lot.

Cheers,
Bart
Logged
Brian Ripley
Newbie
*
Offline Offline

Posts: 4


« Reply #149 on: May 23, 2011, 01:48:45 PM »
ReplyReply

I believe that it is fair to assume that people tend to view A3 sized prints at a greater distance than A4 sized prints. This tendency may not be enough to compensate for the change in print size (i.e. lp/ph may be more important in large images than in small ones).

The human visual system is often said to be limited to 1 minute of arc (angular subtended), and we have a limited ability to focus very near to some object. I believe that this can be used to set a hard limit on how much real resolution is of use for 20/20 vision people not using magnifyers.

Yes, although I think that we can do a little better.  But I am reminded of a friend of mine (a 5 x 4 user) who talked about framed prints coming back from exhibiitions with noseprints on them.  In other words, people will (at least try) to look at very fine details in a large print.

I've done this once.  I was stung at a review of one of my prints (A4 from a Canon 1Ds) to be told I had over-sharpened it.  I asked why he thought so, and was pointed to marks on the edges of the stones in a (600-year old) wall.  I came back the next day with a loupe and asked whether the reviewer had had his eyes tested recently (no ...).  The print resolved the lichen growing on the mortar, but even my younger 6/6 (aka 20/20 in Imperial units) eyes needed a loupe to see all the detail.
Logged
torger
Sr. Member
****
Offline Offline

Posts: 1415


« Reply #150 on: May 25, 2011, 01:35:38 PM »
ReplyReply

I think it is fair to say that a "reasonable" viewing distance is about the same as the width of the picture, with one feet / 30 cm being the closest, that is a postcard is not viewed any closer than an A4. With this viewing distance the full picture can be viewed at once comfortably. Of course, very large panoramic prints may invite the viewer to step close to look at a part of the picture at a time.

I'm just learning about fine art print-making but have done some observations in high quality offset printing. The highest quality offset prints used in photo books swallow 400 ppi. At book-reading distance a 400 ppi photograph with this type of print quality gives the sense that details are smaller than the eye can resolve.

Lowering the ppi to 300 or slightly below makes noticable difference, perhaps not so much that it seems that resolvable detail is lacking but more a sense that surfaces look less natural due to the lack of micro detail. At ~250 ppi the surfaces can look "like plastic".

So this tells me that 400 ppi is probably good, but rather not lower. This is for books at book-reading distance, which can be about A3 in size when opened up, perhaps a bit larger. Let's make it simple and say 24x18 inches, which yields 9600x7200 in 4:3 format which is 69 megapixels.

If you make the print larger, so will the viewing distance (if the viewer is supposed to see the whole image at once), so you will get away with those 69 megapixels regardless of size. More pixels gives you some ability to crop. I think framed pictures behind glass also are a bit less critical, so you can get away with inkjet prints despite that they do not resolve as much as 400 ppi.

This is no exact science of course, but it is clear to me that current DSLRs of ~25 megapixels is a bit too low to match the possible print and eye resolution. However, when DSLRs reach 40 megapixels, which I think they will quite soon (40-50 megapixels is a suitable "limit" for 36x24mm sensor with the current lens resolving power at least) I think many that use medium format today will think DSLRs are "good enough" in terms of image quality. Say with 45 megapixel DSLR you have the same pixel size as current APS-C cameras and you would get 8250x5500, which is 20.6x13.8 inches at 400 ppi (52x35 cm) which would at least be good enough for a spread in books of common sizes. Lowering to inkjet standards were ~300 ppi is considered good enough, those 45 megapixels may be "all that is needed" for any size. But we're not there yet, and still there are advantages of the larger sensor area (gathering more light), and of course there's more to a sensor and readout electronics than only megapixels.
Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1677


« Reply #151 on: May 25, 2011, 02:20:52 PM »
ReplyReply

I think it is wise to separate between camera sensels, image file pixels, and printer dpi. There is no 1:1 correspondence (a 4000x4000 sensel camera will never produce a file with usable information right up to the theoretical limit, a 4800x1200 dpi printer cannot print a 4800x1200 pixel image in a square inch at full spatial/tonal resolution).

Further, the Human Visual System is probably more about acuity than actual detials. As long as an image is sharpened in a pleasing manner, and the printer/display has a pixel/ink shape, size and density that does not detract, we may be able to get by with lower resolution cameras.

-h
Logged
torger
Sr. Member
****
Offline Offline

Posts: 1415


« Reply #152 on: May 25, 2011, 03:17:52 PM »
ReplyReply

I think it is wise to separate between camera sensels, image file pixels, and printer dpi. There is no 1:1 correspondence (a 4000x4000 sensel camera will never produce a file with usable information right up to the theoretical limit, a 4800x1200 dpi printer cannot print a 4800x1200 pixel image in a square inch at full spatial/tonal resolution).

Further, the Human Visual System is probably more about acuity than actual detials. As long as an image is sharpened in a pleasing manner, and the printer/display has a pixel/ink shape, size and density that does not detract, we may be able to get by with lower resolution cameras.

-h

Yes, you're right, and it is very much technology dependent. My discussion on how much megapixels you need is just a rough estimation. A perfectionist wants as high resolution that if increasing it further there's no detectable difference. Some people don't want to include viewing distance either, that is regardless of size they want to max out the printing technology, meaning 300 - 400 ppi regardless of print size.

(When inkjets are involved I think it is important to differ pixels from inkjet raster dots, you need a lot more rastered dpi than ppi to render an image.)

The pixel shape thing is interesting -- personally I think 100% information down to the pixel level in the image file is overrated, probably due to too much pixel-peeping. Actually, some slight fuzziness on the pixel level due to lens resolution limits, diffraction, demosaicing etc is desirable - the sensel size should not be the gating limit of resolution. And the pixel size of the image should preferably slightly outresolve the printing technology at least for close viewing distances.

The thing is that noone likes digital artifacts. While it may be romantic with some film grain in a print noone likes "the jaggies"/pixelation (or any other digital artifact), so you really don't want to present the image in a medium that makes individual pixels visible. This is one reason in that I don't believe in foveon sensors *at all* as long as they are considerably lower resolution than the bayer sensors. The "fuzziness factor" should not be large though (say 15% or so), and it is not in quality bayer systems. I was referring to bayer sensors in the resolution discussion by the way, that is not expecting 100% pixel detail of those 69 megapixels.

It shall be interesting for me to investigate what the upsizing software can do, which promise to repair the pixelation stuff somewhat but I am skeptical. If upsizing software really are fantastic then 100% detail on pixel level might actually be useful, but to me it seems better to increase acutance (sharpen) an image which by nature has just slightly higher pixel resolution than the actual resolution it contains.
Logged
Pages: « 1 ... 6 7 [8]   Top of Page
Print
Jump to:  

Ad
Ad
Ad