Pixel density revisited

Started Oct 22, 2008 | Discussions
Ralf Ronander
Ralf Ronander Contributing Member • Posts: 796
Re: Resampling PS Elements 3.0

I tried downsampling your test target in FastStone, which has several algorithms.
With its Lanszos 3 the result is near perfect, as it is with Bell and Mitchell.
Best however, is BSpline.
All mentioned are better then PSE 3.

DSPographer Senior Member • Posts: 2,464
Re: FastStone Resampling

I tried FastStone and its Lanczos3 resampling is the correct implementation to avoid aliasing except for what it does at the image borders. When you shrink the image to half size horizontally; instead of the first output pixel being located between the first two input pixels, and the last output pixel being located between the last two output pixels; the first and last output pixels are centered on the input pixels. FastStone then spaces the remaining pixel locations to get the desired image size. So to get a spacing of exactly every two input pixels you could for instance scale a 1001 pixel wide image to 501 pixels wide. The problem with this method is that the boundary pixels do not get as much filtering in the downsampling process as the other pixels. This results in a glitch occurring along the right edge of the downsampled in half version of my test image.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
chuxter Forum Pro • Posts: 21,714
But...

attomole wrote:

We view a picture most commonly as a print, how a picture looks under
normal viewing is the most important to me, not what it looks like at
the pixel level, so yes I woulds like to see that bottle label
magnified to the same physical size not number of pixels

When posting the image on the dpr site, how does Phil "magnify to the same physical size"? Unless I'm wrong, which has happened on occasion, a digital image has no physical size. I agree with you, just can't figure out how to do it.

-- hide signature --

Charlie Davis
Nikon 5700, Sony R1, Nikon D300
HomePage: http://www.1derful.info
Bridge Blog: http://www.here-ugo.com/BridgeBlog/
'Experience: Discovering that a claw hammer will bend nails.
Epiphany: Discovering that a claw hammer is two tools...'

 chuxter's gear list:chuxter's gear list
Nikon D810 Nikon D500 Tamron SP 90mm F2.8 Di VC USD 1:1 Macro (F004) Sigma 24-105mm F4 DG OS HSM Tamron SP 150-600mm F/5-6.3 Di VC USD +2 more
bobn2
bobn2 Forum Pro • Posts: 61,424
Re: But...

chuxter wrote:

attomole wrote:

We view a picture most commonly as a print, how a picture looks under
normal viewing is the most important to me, not what it looks like at
the pixel level, so yes I woulds like to see that bottle label
magnified to the same physical size not number of pixels

When posting the image on the dpr site, how does Phil "magnify to the
same physical size"? Unless I'm wrong, which has happened on
occasion, a digital image has no physical size. I agree with you,
just can't figure out how to do it.

An interesting way might be to use a standardised printer driver to render an image of a set size (say A3 for the sake of argument). If it was a Postscript driver, a patch of the 'printer' pixels could be cropped from the file and rendered as a JPEG for viewing on the screen. Obviously the results would depend on the printer driver and JPEG converter selected, but at least it would be the same for every camera, and would lead to a way of estimating the relative output quality between cameras dissimilar in sensor size and/or pixel count.

-- hide signature --

Bob

Iliah Borg Forum Pro • Posts: 25,839
Re: But...

An interesting way might be to use a standardised printer driver to
render an image of a set size

Adobe Acrobat comes to mind However I trust actual scans more.

-- hide signature --
DSPographer Senior Member • Posts: 2,464
Print driver resizing

I don't trust print driver resizing any more than image processing software resizing. What I do when printing is resize in PSE 3 to the native printer resolution before printing; so that resized file or something similar is what I think should be used for reviews. Any software that does an OK resize that minimizes aliasing at least as well as PSE 3 bicubic could become the standard.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
Ralf Ronander
Ralf Ronander Contributing Member • Posts: 796
Re: FastStone Resampling

I´m sure you´re right, but I didn´t say it was perfect. I said it looked better than PSE3, and that Bspline was better still.

J A K Forum Pro • Posts: 15,833
Don't know if you have Qimage Pro but ...

it offers a broad choice of interpolation algorithms for those interested in A versus B comparisons at various print sizes.

Regards,

Joe Kurkjian

Galleries: http://www.pbase.com/jkurkjia

SEARCHING FOR A BETTER SELF PORTRAIT

Ernst Dinkla Regular Member • Posts: 293
Re: Don't know if you have Qimage Pro but ...

J A K wrote:

it offers a broad choice of interpolation algorithms for those
interested in A versus B comparisons at various print sizes.

Like I mentioned. On down-sampling it also has an adjustable anti-aliasing filter.

To create TIFF's one can use the print to file function and have the same algorithms in the process.

Ernst Dinkla

bobn2
bobn2 Forum Pro • Posts: 61,424
Re: But...

Iliah Borg wrote:

An interesting way might be to use a standardised printer driver to
render an image of a set size

Adobe Acrobat comes to mind However I trust actual scans more.

Yes, I think I agree, but the method I proposed is more readily automatable (in case anyone was to argue that printing and scanning was too much effort), less prone to human error and to arguments about the quality of the scanner. I think I'd want to go for something open source rather than proprietary, it means that should there be arguments about the nature of the resizing process, the code's all there for anyone who cares to examine.

-- hide signature --

Bob

bobn2
bobn2 Forum Pro • Posts: 61,424
Re: Print driver resizing

DSPographer wrote:

I don't trust print driver resizing any more than image processing
software resizing. What I do when printing is resize in PSE 3 to the
native printer resolution before printing; so that resized file or
something similar is what I think should be used for reviews. Any
software that does an OK resize that minimizes aliasing at least as
well as PSE 3 bicubic could become the standard.

The point is not what is trusted, but about making a meaningful comparison as to the probable output quality of different cameras. I would think whatever driver was selected, it would need to be open source, so that its properties are understood and open to public discussion.
The 'known algorithm resize' is good, too.
--
Bob

ejmartin Veteran Member • Posts: 6,274
Re: Print driver resizing

bobn2 wrote:

DSPographer wrote:

I don't trust print driver resizing any more than image processing
software resizing. What I do when printing is resize in PSE 3 to the
native printer resolution before printing; so that resized file or
something similar is what I think should be used for reviews. Any
software that does an OK resize that minimizes aliasing at least as
well as PSE 3 bicubic could become the standard.

The point is not what is trusted, but about making a meaningful
comparison as to the probable output quality of different cameras. I
would think whatever driver was selected, it would need to be open
source, so that its properties are understood and open to public
discussion.
The 'known algorithm resize' is good, too.

Indeed. One might for instance adopt ImageMagick; open source, with an implementation of the standard interpolation filters and user selectable parameters.

On the other hand, "our proprietary testing software" carries a certain cachet and mystique

-- hide signature --
DSPographer Senior Member • Posts: 2,464
ImageMagick resizing

I looked at ImageMagick to convert an image to half size and while it seems to correctly treat the image edges it looks like it was only using a two lobe window instead of the three lobe Lanczos window that the documentation said it would. It still worked fairly well I am just wondering why the default parameters seemed different. It could be operator error though, using a command line interface seems like going back 15 years.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
DSPographer Senior Member • Posts: 2,464
Re: ImageMagick resizing

If I specify the resize method with the "-filter lanczos" option then ImageMagick does use the Lanczos a=3 algorithm. I don't know what method it is using by default but it has more aliasing then Lanczos3.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
rhlpetrus Forum Pro • Posts: 25,860
Re: not sure if pixel density but A900

Iliah Borg wrote:

rhlpetrus wrote:

Iliah Borg wrote:

I do not think it is the right thing to do - to compare OOC jpegs and
make a verdict on sensor. In-camera jpeg engine can be improved with
firmware updates, by the way. But if you are comparing OOC jpegs in
regards to noise, compare resolution too, and not on a
black-and-white target.

Not a verdict, just but a bad omen

The DR which is claimed to be good is also the jpeg test from Dpr,

And do you know what are the results of DPR and Seconic tests of OOC
jpegs on 5D original? You may be amazed by the difference.

?

I have A900, and I'm completely satisfied with this camera for my
intended use. I would be very happy to have that sensor in
D-one-digit body because the lens choice with Sony is quite limited
and with the wider choice I would use that camera much more.

Hey, I'm happy with a D80 ;)!! Now, I think the 5DII will be, at least until Nikon shows its stuff, really better than the A900 re IQ. And C lenses are pretty good as well. But I'm sure the A900 in good light will give superb results. Only the Hassy x A900 test posted in Sony's forum really makes one wonder if FF 35mm is the way re high MP cameras for landscapes. Maybe that's why all this talk about MX ... (which, btw, I find likely to be a prank).
--
Regards, Renato.
http://www.flickr.com/photos/rhlpedrosa/
OnExposure member
http://www.onexposure.net/

You cannot depend on your eyes when your imagination is out of focus
(Mark Twain)

 rhlpetrus's gear list:rhlpetrus's gear list
Leica D-Lux (Typ 109) Nikon D7000 Nikon 1 V1 Nikon AF-S DX Nikkor 12-24mm f/4G ED-IF Nikon AF Nikkor 35mm f/2D +4 more
chuxter Forum Pro • Posts: 21,714
Re: Print driver resizing

DSPographer wrote:

I don't trust print driver resizing any more than image processing
software resizing.

I don't think "trust" is the right word? My issue is that with most printer drivers, we don't know HOW the print is resized.

What I do when printing is resize in PSE 3 to the
native printer resolution before printing; so that resized file or
something similar is what I think should be used for reviews. Any
software that does an OK resize that minimizes aliasing at least as
well as PSE 3 bicubic could become the standard.

Yes, that's a good approach, but still I am not sure how to crop and present a same-sized enlargement of various camera outputs.

Let's go through the exercise and see how far I get...

I'd like to standardize on a large 2:3 enlargement. Since many people have 13" x 19" printers (A3+ size) I think a 12" x 18" enlargement should be the "standard".

I'd suggest we use 300 dpi as the standard resolution.

SO, let me take the 3 cameras I have access to and see how I would use PSE to create a small crop from each of them that looks like a part of a 12" x 18" enlargement. I'll start by defining the small crop to be 4" x 4"...2 of those will easily fit on a computer monitor so we can do A-B comparisons. Doing the math, that's 1200 x 1200 px for the crop.

1) The Nikon CP5700: 2560 x 1920 px => to get a 12" x 18" print we'll have to print at 160 dpi => a 4" x 4" crop will be 640 x 640 px =>

2) The Sony R1: 3888 x 2592 px => to get a 12" x 18" print, we'll have to print at 216 dpi => a 4" x 4" crop will be 864 x 864 px =>

3) The Nikon D300: 4288 x 2848 px=> to get a 12" x 18" print, we'll have to print at 237 dpi => a 4" x 4" crop will be 948 x 948 px =>

It's easy to create these crops. The problem is that when we try to display them on a monitor, they are not the same size:

As you can see, that 948 px high strip is much too big and besides how can we compare the IQ of the 3 cameras if the crops are different sizes.

OK, you say...first normalize the sizes by using PSE...then reduce the overall size by reducing everything by 50%. That works, but it seems to be testing the characteristics of the PSE resample engine more than anything else! Is there any data to show that PSE (or any other editor) doesn't introduce small errors when taking 3 pix and resampling them at 50%m 55%, and 74%? With this uncertainty, how can we ever truly compare apples to apples???

-- hide signature --

Charlie Davis
Nikon 5700, Sony R1, Nikon D300
HomePage: http://www.1derful.info
Bridge Blog: http://www.here-ugo.com/BridgeBlog/
'Experience: Discovering that a claw hammer will bend nails.
Epiphany: Discovering that a claw hammer is two tools...'

 chuxter's gear list:chuxter's gear list
Nikon D810 Nikon D500 Tamron SP 90mm F2.8 Di VC USD 1:1 Macro (F004) Sigma 24-105mm F4 DG OS HSM Tamron SP 150-600mm F/5-6.3 Di VC USD +2 more
DSPographer Senior Member • Posts: 2,464
Re: Print driver resizing

chuxter wrote:

OK, you say...first normalize the sizes by using PSE...then reduce
the overall size by reducing everything by 50%. That works, but it
seems to be testing the characteristics of the PSE resample engine
more than anything else! Is there any data to show that PSE (or any
other editor) doesn't introduce small errors when taking 3 pix and
resampling them at 50%m 55%, and 74%? With this uncertainty, how can
we ever truly compare apples to apples???

I think we are trying to answer the same question: how to compare apples to apples when starting from different image sizes? Even when comparing prints the images will have been "processed": the print driver will typically do some interpolation; then when viewing the image depending on viewing distance your eyes will average over a small area of the picture. In my opinion a high quality resizing algorithm can recreate this effect in a well controlled way. As long as the algorithm and standard image sizes are carefully chosen the ranking of two cameras should be affected less by resize algorithm errors than by such things as print driver settings like output sharpening amount.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
bobn2
bobn2 Forum Pro • Posts: 61,424
Re: Print driver resizing

DSPographer wrote:

chuxter wrote:

OK, you say...first normalize the sizes by using PSE...then reduce
the overall size by reducing everything by 50%. That works, but it
seems to be testing the characteristics of the PSE resample engine
more than anything else! Is there any data to show that PSE (or any
other editor) doesn't introduce small errors when taking 3 pix and
resampling them at 50%m 55%, and 74%? With this uncertainty, how can
we ever truly compare apples to apples???

I think we are trying to answer the same question: how to compare
apples to apples when starting from different image sizes? Even when
comparing prints the images will have been "processed": the print
driver will typically do some interpolation; then when viewing the
image depending on viewing distance your eyes will average over a
small area of the picture. In my opinion a high quality resizing
algorithm can recreate this effect in a well controlled way. As long
as the algorithm and standard image sizes are carefully chosen the
ranking of two cameras should be affected less by resize algorithm
errors than by such things as print driver settings like output
sharpening amount.

No test is going to give a completely accurate indication of the results any particular photographer will get form their workflow. Short of producing outputs for every permutation of tool and parameter setting, there is no way of doing it. Morever, when you get a new camera, you adjust your workflow to get the best results for you. What is needed is a test with enough supporting information for an informed user to make a judgement of how it would work for them, and an 'at a glance' usability that isn't going to mislead the uninformed user. The trouble is with the 100% crop is that it meets neither of those objectives. The resized crop will do both. At a glance it will give an indication of likely relative output quality, with the normal caveats. For the more informed, after gaining first impressions, if you know the workflow behind the test, you can make a judgment on how that relates to your own.
--
Bob

chuxter Forum Pro • Posts: 21,714
Re: Print driver resizing

Yes, I agree that our goals are the same, but you didn't specifically comment on my question:

"Is there any data to show that PSE (or any other editor) doesn't introduce small errors when taking 3 pix and resampling them at 50%m 55%, and 74%?"

Perhaps a rewording would make the issues (in my mind) clearer? How about:

"Is there data to quantify how various resampling algorithms work when asked to do slightly different reduction ratios?" I assume that exactly 50% is a sweet-spot...

DSPographer wrote:

chuxter wrote:

OK, you say...first normalize the sizes by using PSE...then reduce
the overall size by reducing everything by 50%. That works, but it
seems to be testing the characteristics of the PSE resample engine
more than anything else! Is there any data to show that PSE (or any
other editor) doesn't introduce small errors when taking 3 pix and
resampling them at 50%m 55%, and 74%? With this uncertainty, how can
we ever truly compare apples to apples???

I think we are trying to answer the same question: how to compare
apples to apples when starting from different image sizes? Even when
comparing prints the images will have been "processed": the print
driver will typically do some interpolation; then when viewing the
image depending on viewing distance your eyes will average over a
small area of the picture. In my opinion a high quality resizing
algorithm can recreate this effect in a well controlled way. As long
as the algorithm and standard image sizes are carefully chosen the
ranking of two cameras should be affected less by resize algorithm
errors than by such things as print driver settings like output
sharpening amount.

-- hide signature --

Charlie Davis
Nikon 5700, Sony R1, Nikon D300
HomePage: http://www.1derful.info
Bridge Blog: http://www.here-ugo.com/BridgeBlog/
'Experience: Discovering that a claw hammer will bend nails.
Epiphany: Discovering that a claw hammer is two tools...'

 chuxter's gear list:chuxter's gear list
Nikon D810 Nikon D500 Tamron SP 90mm F2.8 Di VC USD 1:1 Macro (F004) Sigma 24-105mm F4 DG OS HSM Tamron SP 150-600mm F/5-6.3 Di VC USD +2 more
DSPographer Senior Member • Posts: 2,464
Re: Print driver resizing

Yes there is data on how resizing to various ratios works. By using test images like the sweep test and rings test (zone plate) as well as ordinary photos the results of various resizing ratios can be tested. For special ratios like 1/2 the fractional location of the output point may stay fixed but that doesn't necessarily mean the result will be any better than some arbitrary ratio. It turns out that for down-sampling with anti-aliasing filters the exact ratio doesn't usually matter much: the filtering step generally results in the quality of all the fractional output point positions being just about the same so a 57% resizing would look just as good as 50%. This is different than the up-sizing case where the intent is to leave original points unfiltered- in that case exact ratios like 2:1 up-sizing typically yield the best results.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
Keyboard shortcuts:
FForum MMy threads