Upsizing with LR or Perfect Resize?

[Bart_van_der_Wolf]

I strongly believe that the only credible way of argumentation has to be reproducible/verifiable by independent others. Just like in proper science, full disclosure and peer review is the way forward. Just stating opinions doesn't benefit anybody.

I guess I'll have to prepare yet again some sample material (unless somebody beats me to it) that's freely available to whoever wants to do their verification of what can be achieved. I'll need a bit of time to create such a target file that would ideally have subject matter that's diverse enough to suit many different types of users. The original should also be small enough to allow more significant upsampling without producing output files that are too large to store comfortably or download.

As a first attempt of a target with multiple diverse subject crops, to give some homework for the weekend ahead, I've downsampled an image with lots of detail to 30% of it's original pixel size. To save a bit of space for future additions, it is saved as a JPEG with Photoshop at maximum quality. You can download it here.

Try upsampling (a crop) 3x, 5x, 10x, with Lightroom and Perfect Resize. Other applications can be used, but remember that the thread is specifically about the two mentioned. You can post either comparison crops, for scrutiny on display (e.g. at output size zoom, if the PPI was set to your printer's optimum), or even better print them side by side (also better used ink than with an unclog operation).

At 3x upsample, the total image becomes 4194 x 5235 pixels or 11.65 x 14.54 inches at 360 PPI (13.98x17.45 inch @ 300 PPI). The differences are beginning to show, not very obvious in print, but there is a bit more bite in the PR version.
You can also upsample to 6x the pixel dimensions (8388 x 10470 pixels) for printing at 720 PPI, which will benefit PR which will add edge detail with even more resolution.

Do not immediately dismiss the possibility to print at 720 PPI by doubling the output's pixel dimensions, even if the image is small and already needs to be upsampled to reach 360 PPI, it shows in output quality when the upsampling application adds edge resolution.

At 5x upsample, the total image becomes 6990 x 8725 pixels or 19.42 x 24.24 inches at 360 PPI (23.30x29.08 inch @ 300 PPI).
At 10x upsample, the total image becomes 13980 x 17450 pixels or 38.83 x 48.47 inches at 360 PPI (46.60x58.17 inch @ 300 PPI). We're starting to get into the Wall high Booth displays territory. At these sizes, and their 2x larger versions for higher PPI printing, the differences become even more pronounced. And of course, it's better to have more original pixels to begin with, but that's a different discussion.

Maybe it is possible to make something that at least looks somewhat like a silk purse, from a sow's ear afterall ... Not dropping any quality/stitches on the way may also help ...

Cheers,
Bart

[TonyW]

Hi Bart
Interesting idea and thanks for posting.  I am one of those that had used both GF and PR and felt that there was little benefit if any and came to the conclusion that I actually preferred the results from PS which I found to be marginally softer but seemed to hold detail rather than become blocky.  

I actually asked the question on another forum about this as it seemed to me that prior to PS CS5 GF offered better quality but my testing of PR with CS5 showed little advantage.  
Out of curiosity I revived my version of PR and ran a couple of tests using PR, LR and CS6 at 10x 300ppi and (due to a calculating error on my part!!) 41+x 600 ppi.  

I had not realised the passing of time and my last look at this was in 2010!  Time flies and software no doubt has improved I am not sure what version PR is at but mine is v7.0 pro.

Under the circumstances and assuming you are using the latest version I did not think it appropriate to post - I will if you are interested.

FWIW  LR 5.5 seems to produce an image with better resolved detail than CS6 whereas PR 7 has the blocky (fractal look!).  I have not actually bothered to print out the images (and I do know that this is important, printer out of service) but it is interesting to view on screen at editing distance where it is obvious that LR version is very soft but still retains detail and the PR image looks awful.  

Stand back from the screen of course and the PR image improves to the point where it does appear to be better, but there are so many variables.  For instance how close are we going to view the print, photographers distance or normal persons distance?  I think for the former although the image is going to look very soft may be preferable.

Has the latest version of PR improved significantly since version 7

[Bart_van_der_Wolf]

Hi Tony,

Indeed, resizing in LR is better than what Photoshop (at least till CS6) can achieve.

Under the circumstances and assuming you are using the latest version I did not think it appropriate to post - I will if you are interested.

I am preparing a comparison, as we speak/write.

FWIW  LR 5.5 seems to produce an image with better resolved detail than CS6 whereas PR 7 has the blocky (fractal look!).  I have not actually bothered to print out the images (and I do know that this is important, printer out of service) but it is interesting to view on screen at editing distance where it is obvious that LR version is very soft but still retains detail and the PR image looks awful.  

Stand back from the screen of course and the PR image improves to the point where it does appear to be better, but there are so many variables.  For instance how close are we going to view the print, photographers distance or normal persons distance?  I think for the former although the image is going to look very soft may be preferable.

The difficulty is that one needs to (pre-)view the result at the correct zoom factor, in order to get a close to identical magnification as the final output will have. The somewhat artificial looking PR image at 100% pixel peeping zoomed size with a relatively low display PPI will then start looking like what one can expect in print. The blocky look (which may also be tuned with the various PR parameters) will then disappear at normal viewing distances. Should close-up scrutiny be allowed, then one can adjust the setting to account for that.

The display  zoom down-sampling implementation in Photoshop is rather crude, but it will be somewhat useful nonetheless. One needs to either setup the screen PPI in the Photoshop preferences for the View>Print Size menu to do the right thing, or just type the correct zoom factor in the window's zoom status in the lower left corner. It assumes that the document has the correct output PPI tag settings. The document's PPI tag can be changed by using the Image>Image Size... menu, with the resample option switched off.

When I work on my laptop, with a 128 PPI display (just measure the display's width in inches and divide the number of display pixels by that width for your screen), i need to use screen / output PPI = zoom, e.g. 128 / 360 = 0.35555 or 35.56%.

Has the latest version of PR improved significantly since version 7

I'm not sure. I wasn't able to run PR7 because it kept crashing, but Perfect Resize Version 8.1 runs fine on the same setup. I'll post the settings with my examples, maybe they are identical for both version 7 and 8.

Cheers,
Bart

P.S. I've already attached a suggestion for the PR settings for a 3x upsample, in order to compare if PR7 has the same parameters that can be adjusted.

[Bart_van_der_Wolf]

As a first attempt of a target with multiple diverse subject crops, to give some homework for the weekend ahead, I've downsampled an image with lots of detail to 30% of it's original pixel size. To save a bit of space for future additions, it is saved as a JPEG with Photoshop at maximum quality. You can download it here.

Try upsampling (a crop) 3x, 5x, 10x, with Lightroom and Perfect Resize. Other applications can be used, but remember that the thread is specifically about the two mentioned. You can post either comparison crops, for scrutiny on display (e.g. at output size zoom, if the PPI was set to your printer's optimum), or even better print them side by side (also better used ink than with an unclog operation).

To kick-off, here is a crop of a 3x upsample of the linked JPEG in the form of an animation (download and display in an image viewer or browser, without resizing/zooming for a 100% zoom view, i.e. pixel for pixel peeping). To get an impression of how it looks in print, zoom-out to a print size equivalent zoom setting.

Divide your display PPI (my laptop is 128 PPI) by 360 PPI if you want to simulate print size. You'll see that what may look ugly at 100% zoom with very low resolution display dots, is very nice at (in my case 128/360 = 35.56%) the proper zoom setting. The only reason I added the Photozoom example is to demonstrate the difference between 100% zoom and print size adjusted zoom, not to specifically address Photozoom in this thread. All samples could be further altered by e.g. Topaz Detail, and noise can be added to give more structure to smooth areas.

I'll later (done, as attached) add the settings that were used (Perfect Resize was set as posted earlier), but those are of course also subject to personal preference and output medium. Anyway, Lightroom was set to Glossy Output Standard sharpening, the others were sharpened only by the Resampling application's settings for sharpening.

Cheers,
Bart

P.S. Added attachments with the settings that were used.

[TonyW]

The difficulty is that one needs to (pre-)view the result at the correct zoom factor, in order to get a close to identical magnification as the final output will have. The somewhat artificial looking PR image at 100% pixel peeping zoomed size with a relatively low display PPI will then start looking like what one can expect in print. The blocky look (which may also be tuned with the various PR parameters) will then disappear at normal viewing distances. Should close-up scrutiny be allowed, then one can adjust the setting to account for that.

The display  zoom down-sampling implementation in Photoshop is rather crude, but it will be somewhat useful nonetheless. One needs to either setup the screen PPI in the Photoshop preferences for the View>Print Size menu to do the right thing, or just type the correct zoom factor in the window's zoom status in the lower left corner. It assumes that the document has the correct output PPI tag settings. The document's PPI tag can be changed by using the Image>Image Size... menu, with the resample option switched off.

When I work on my laptop, with a 128 PPI display (just measure the display's width in inches and divide the number of display pixels by that width for your screen), i need to use screen / output PPI = zoom, e.g. 128 / 360 = 0.35555 or 35.56%.

I'm not sure. I wasn't able to run PR7 because it kept crashing, but Perfect Resize Version 8.1 runs fine on the same setup. I'll post the settings with my examples, maybe they are identical for both version 7 and 8.

Cheers,
Bart

P.S. I've already attached a suggestion for the PR settings for a 3x upsample, in order to compare if PR7 has the same parameters that can be adjusted.

Hi Bart,
I was aware of the problems you pointed out and in another post mentioned that I do not think it of any benefit trying to evaluate a 1:1 view in PS vs actual print size as for me with a 100 ppi monitor and printing normally to a Canon or HP both requiring 300/600 ppi.  The 33.33% / 16.67% reduction to achieve something like viewing actual print size is not at all helpful in evaluating image quality at least that is my experience.

So in this case perhaps little real world value unless the finished image is to be presented on screen at a 3-6x loupe view?

When setting the zoom to 33.33% images look a lot closer and side by side without any image massaging in CS6 in some ways, and for certain subjects PR may score a little higher in perceived resolution/sharpness.  I have found with the application of a curves layer and a sharpening layer in PS that I generally still like the PS version a little better and although it has been a long time also found that the same held for printing.  I really do think that image content plays a role here as there is a tendency for diagonals and circles to display 'jaggies' in PS or LR whereas PR smooths them out somewhat.  Your example image actually quite good as in LR the rope actually looks like rope I think due to this artefacting .  But my impressions are that for more organic subjects faces etc the PS versions preferable.

Have attached a screenshot next to yours of the v7 options and although very similar I do not that you have a choice for Gen. Fract. and maybe more not available in this older version.  I think you would likely agree that it would be unwise to assume nothing much had changed between versions in 4 years  .  Therefore safer to assume that all applications have matured and output quality improved in this time?  I will try a 3x with your settings though

One thing that I did find incredibly annoying is the time required by PR to make changes once you altered a single parameter for resizing or sharpening and also the long time to save and write a JPEG image.  For instance to save a 280x339 pixel crop @300ppi from your image in LR or PS around 3-5 seconds and PR around 40 seconds.  Larger image took nearly 1/2 hour PR vs 20 seconds or so in PS.  Still this is a system shortcoming I think rather than the applications as my current PC is coal fired and at over 5 years old an antique by today's standards  

[TonyW]

...I'm not sure. I wasn't able to run PR7 because it kept crashing, but Perfect Resize Version 8.1 runs fine on the same setup. I'll post the settings with my examples, maybe they are identical for both version 7 and 8.

Cheers,
Bart

P.S. I've already attached a suggestion for the PR settings for a 3x upsample, in order to compare if PR7 has the same parameters that can be adjusted.

Hi Bart
I don't think that PR7 will run as a plugin in CS6 at least on my system (Win 7) I found out that the standalone version still there and works but very clunky.

Just an update FWIW I did the 3x upsample with the same parameters that you showed in your screenshot and viewing with mine overlayed on yours I can see differences (at 100%) not huge but nevertheless observable but...  As mine is a JPEG and yours GIF and not knowing what parameters were used for the GIF I would not want to try and make a judgement. Your gif looked sharper exhibited a little grain but also appeared to smooth the jaggies around the ropes and rigging.  My assumption is that at least some of the difference is merely due to the formats

[Bart_van_der_Wolf]

Hi Bart
I don't think that PR7 will run as a plugin in CS6 at least on my system (Win 7) I found out that the standalone version still there and works but very clunky.

Hi Tony,

Photoshop filters are not capable of resizing image dimensions, so the action is probably under File>Automate>Perfect Resize X...

Just an update FWIW I did the 3x upsample with the same parameters that you showed in your screenshot and viewing with mine overlayed on yours I can see differences (at 100%) not huge but nevertheless observable but...  As mine is a JPEG and yours GIF and not knowing what parameters were used for the GIF I would not want to try and make a judgement. Your gif looked sharper exhibited a little grain but also appeared to smooth the jaggies around the ropes and rigging.  My assumption is that at least some of the difference is merely due to the formats

Most likely caused by the different file formats, and probably not that huge/important when zoomed out to output dimensions. I have no information about PR version 7 vs version 8 differences, but it sounds like they are similar.

Cheers,
Bart

[TonyW]

Thanks Bart, decided to RTFM and found the reason that PR 7 does not show in CS6 is that it requires update to 7.0.7.  Once done all as it should be.

[Bart_van_der_Wolf]

Thanks Bart, decided to RTFM and found the reason that PR 7 does not show in CS6 is that it requires update to 7.0.7.  Once done all as it should be.

Hi Tony,

That's probably the reason I couldn't get it to run either, after which I gave up until a reduced price offer crossed my path with version 8.

Cheers,
Bart

[TonyW]

Hi Bart,
As yet not a great take up so I thought I would add a couple of quick plays.  Now these are screen grabs using sniping tool and Save for Web via CS6.  

Not the best way I know but it does at least show a comparison as seen on screen and if I have done all correctly the 33.33% and the 16.67% view should display sized correctly and at 100% view the inch marks could be checked to see if close enough.

Method:
Cropped your image to 280x339 pixels and imported into LR

For the 10x image in LR exported image sized to 3390 pixels long side, sharpen for print standard, glossy paper.

For the 10x PR7 used the settings you provided earlier, however I now realise that these were meant for the 3x upsample so are probably not optimal.

I mentioned earlier that I had a technical hiccup (proper name cock up!) and managed to size the cropped image to 11,667 x 14125 pixels around 41.62x, so include that as well.

Hopefully it should be apparent what I have done viewing the images.  For some reason I decided to see what would happen to the LR image when sharpened just using USM in CS6 and these are also shown.

I am not able to print at the moment and I find it very difficult to predict how good either of these images would look and if one better than the other either at normal viewing distance or much closer.  My gut feeling is that I would probably prefer the LR/PS version with some appropriate sharpening when viewed close up but this may be biased based on previous experience

[Bart_van_der_Wolf]

Hi Bart,
As yet not a great take up so I thought I would add a couple of quick plays.  Now these are screen grabs using sniping tool and Save for Web via CS6.  

Not the best way I know but it does at least show a comparison as seen on screen and if I have done all correctly the 33.33% and the 16.67% view should display sized correctly and at 100% view the inch marks could be checked to see if close enough.

Hi Tony,

Yes, at 100% view the inches are approx. correct, so this should give a decent first impression of how actual output might look and how close one can get before things start looking odd.

Method:
Cropped your image to 280x339 pixels and imported into LR

For the 10x image in LR exported image sized to 3390 pixels long side, sharpen for print standard, glossy paper.

For the 10x PR7 used the settings you provided earlier, however I now realise that these were meant for the 3x upsample so are probably not optimal.

There is also personal preference involved, so correct settings do not really exist. The 3x settings are usable, although the higher the output magnification the more one may reduce the edge enhancement, to avoid what I tend to call 'a mental disconnect between the amount of edge sharpness an surface/texture detail'.

The results with these settings would be usable. They also make clear IMHO that LR output at these magnifications leaves a lot to be desired, and is certainly not 'better'.

Hopefully it should be apparent what I have done viewing the images.  For some reason I decided to see what would happen to the LR image when sharpened just using USM in CS6 and these are also shown.

Yes, that is always an option. In fact I would recommend re-sharpening the LR output if one only has LR available and not PR. Focus Magic can be useful, if  computer memory requirements are not a problem, but even a Smart sharpen deconvolution would work to some extent if one doesn't have Focus Magic.

I am not able to print at the moment and I find it very difficult to predict how good either of these images would look and if one better than the other either at normal viewing distance or much closer.  My gut feeling is that I would probably prefer the LR/PS version with some appropriate sharpening when viewed close up but this may be biased based on previous experience.

That would be the ultimate proof of the pudding, so to speak. You may learn to appreciate the PR result as well, depending on subject matter and input file quality.

Thanks for doing the tests yourself, I feel less lonely ... Your tests confirm what I see, and indeed good post-upsample sharpening of the LR output can even improve those results further (which shows that they are not optimal out of the box).

Cheers,
Bart

[Nora_nor]

I remember  year or two years ago I had to print some files for an exhibition (photo club) (30x40cm), and they could not find the original files just then (had put all the image files on external drives because of full HD) and only had web images. I had perfect resize trial on one mac and the prints turned out just fine! It would not have made any sense to try LR, so I do not have anything to compare with sadly.

[eronald]

Bart,

With all due respect, I don't think that Jpeg -a destructive transform even at max qual- helps your case.
Clean data is usually a good way to start.

Edmund

As a first attempt of a target with multiple diverse subject crops, to give some homework for the weekend ahead, I've downsampled an image with lots of detail to 30% of it's original pixel size. To save a bit of space for future additions, it is saved as a JPEG with Photoshop at maximum quality. You can download it here.

Try upsampling (a crop) 3x, 5x, 10x, with Lightroom and Perfect Resize. Other applications can be used, but remember that the thread is specifically about the two mentioned. You can post either comparison crops, for scrutiny on display (e.g. at output size zoom, if the PPI was set to your printer's optimum), or even better print them side by side (also better used ink than with an unclog operation).

At 3x upsample, the total image becomes 4194 x 5235 pixels or 11.65 x 14.54 inches at 360 PPI (13.98x17.45 inch @ 300 PPI). The differences are beginning to show, not very obvious in print, but there is a bit more bite in the PR version.
You can also upsample to 6x the pixel dimensions (8388 x 10470 pixels) for printing at 720 PPI, which will benefit PR which will add edge detail with even more resolution.

Do not immediately dismiss the possibility to print at 720 PPI by doubling the output's pixel dimensions, even if the image is small and already needs to be upsampled to reach 360 PPI, it shows in output quality when the upsampling application adds edge resolution.

At 5x upsample, the total image becomes 6990 x 8725 pixels or 19.42 x 24.24 inches at 360 PPI (23.30x29.08 inch @ 300 PPI).
At 10x upsample, the total image becomes 13980 x 17450 pixels or 38.83 x 48.47 inches at 360 PPI (46.60x58.17 inch @ 300 PPI). We're starting to get into the Wall high Booth displays territory. At these sizes, and their 2x larger versions for higher PPI printing, the differences become even more pronounced. And of course, it's better to have more original pixels to begin with, but that's a different discussion.

Maybe it is possible to make something that at least looks somewhat like a silk purse, from a sow's ear afterall ... Not dropping any quality/stitches on the way may also help ...

Cheers,
Bart

[Bart_van_der_Wolf]

Bart,

With all due respect, I don't think that Jpeg -a destructive transform even at max qual- helps your case.
Clean data is usually a good way to start.

Hi Edmund,

All serious comments are appreciated, yours included.

Actually, you'd be surprised if you did a Difference layer comparison between a maximum quality JPEG conversion, and a TIFF source, both at 8-b/ch (see attached histogram panel for the statistical summary of this particular image). Also zooming in to the image itself, the vast majority of pixels (93.26% of the pixels are identical or 1 different from the original 2439510 pixels, 99.20% is no more than 2 different) are perceptually indistinguishable from the original. Besides, upsamplng is upsampling even if it had to be done based on poor input quality images, unlike this image.

BTW, 100% quality JPEGs are excellent for off-site printing, because of the (especially aimed at perceptually) small differences, assuming they were already converted to the destination output profile.

Cheers,
Bart

[digitaldog]

Actually, you'd be surprised if you did a Difference layer comparison between a maximum quality JPEG conversion, and a TIFF source, both at 8-b/ch (see attached histogram panel for the statistical summary of this particular image)

Just did a similar test using a Granger Rainbow created in 16-bit Lab initially. So a large gamut working space. I usedApply Imagecommand/subtract which should produce the same results as using layers and my Std Dev is even lower than yours (0.57). I hear what Edmund is suggesting using cleaner data but the JPEG IMHO doesn't appear to be a factor and would agree, visually the differences are totally insignificant.

The tests were saved as ProPhoto RGB too, something I'd not recommend as best practice for a JPEG.

[JRSmit]

So in essence what conclusion can be drawn? Is 16 bit tiff in prophotorgb an overkill? Can we just stick to 8bit jpeg in srgb?

[TonyW]

Hi Bart
Thought I should actually print and compare  

My crop of your image brought into CS6 and a duplicate of the background image made to serve as a sharpening layer.  Smart sharpen Lens blur used to get as close to the PR result as possible.

I printed out the 10x 300 ppi crop PR and LR side by side.  Only printer available to me at this time a cheap (but far from nasty) Canon A4.  Having no suitable Canon paper I used HP semi gloss, and because I do not try and use this printer for colour critical work it uses third party inks.  Set for highest quality and through CS6 I let printer manage colour at Canons own semi gloss paper settings.

Scanned at 300 dpi using Epson v500 with all sharpening etc turned off.

FWIW attached is the full size scan of the print further cropped to file size acceptable by the forum.

My impressions viewing the original – just in case my attachment does not appear as it should (EDIT:had to link to Photobucket as I think exceeded forum size limits):

The original image if resized 10x and printed at 300ppi becomes 46.6” x 58.17”

Accepting a ‘normal’ viewing distance of 1.5x print diagonal gives a distance from print of 9’.  I can see no difference at this range.
At around 3’ I can see some differences.  In fact the PS version appears to be marginally better at resolving detail in the figures and carvings with no discernible difference to the curved, horizontal or diagonals.

At 12” the differences more apparent.  PR7 shows improvement in edge sharpness without introducing jaggies that are more apparent in the PS sharpened version.  But IMHO loses out in the ability to retain detail again particularly apparent in the carvings although not exactly sharp the impression remains of more detail.

My thoughts/conclusions:

I had a yardstick to measure against and produce an image as close as I could to the PR7 using the suggested settings.

I think it is difficult to do a really fair comparison between the products.  Using the PR settings that you provided for the 3x image on the 10x version yielded sharper edges but loss of image detail therefore would need to be toned down a little IMO in an attempt to recover that detail if that was the important criteria for this particular image.

My editing tool of choice for this type of work is PS and cannot think of a way to match the sharpening effect in this particular PR7 example in LR alone.
 
Of course the PR version could also be massaged in many ways within another editing or sharpening application.

I think it would take many tests with wildly varying image content to try and come to any valid generalisation about the relative merits of one vs the other and I strongly suspect that the conclusion likely to be that image content plays a vital role as does the individuals idea of what constitutes ‘best’ image quality

EDIT: Note this image has been scaled by Photobucket but is reasonable representation

[digitaldog]

So in essence what conclusion can be drawn? Is 16 bit tiff in prophotorgb an overkill? Can we just stick to 8bit jpeg in srgb?

Absolutely not! The differences you're seeing in my example and I believe Bart's is only due to the file format saved, that from a TIFF vs. a JPEG saved at highest quality (12). The image had no editing!

[David Eckels]

Tony,
But which is which?

[TonyW]

Tony,
But which is which?

Sorry, LR/CS6 sharpened image left.  If you click on the screen image it links to Photobucket and if you then click the magnify icon you will get the 100% view