Interpolating in Lightroom 2 vs. Photoshop

[httivals]

I just read in Chris Orwig's Lightroom 2 book that enlargening an image through Lightroom's "enlarge" export option gives superior results to interpolating up in Photoshop.  This is the first I've heard of that.  Is this correct?  I've experimented quite a bit in printing 18" x 24" images from my Canon 5D.  What I've found works best thus far is (1) do capture sharpening in DXO Optics; (2) other corrections in Lightroom, then export to photoshop with no other sharpening; (3) uprez in photoshop using bicubic smoother to 360 dpi to print at 24" wide; (4) add a bit of grain using Photokit Sharpener -- usually at the 200 or 400 film simulation setting; (5) output sharpening using Photokit Sharpener; (6) Print.  (By the way I've also experimented with genuine fractals, photozoom pro, and after many tests determined that the above workflow works best for me.)

I realize that uprezzing and sharpening varies by individual tastes.  However, it makes sense to me that Lightroom could use a superior enlargening algorithim because there's more data available when working on RAW vs. a TIFF or PSD file.  Is this correct that Lightroom's uprezzing algorithim is different than Photoshop's?

[madmanchan]

Yes, LR's resampling algorithm is different from Photoshop's, with different strengths and weaknesses.

[httivals]

Yes, LR's resampling algorithm is different from Photoshop's, with different strengths and weaknesses.

Eric: are the strengths and weaknesses of LR's resampling algorithm documented somehwere, in particular with respect to up-sampling?  Or, do you have a recommendation as to which is better for uprezzing in the example/workflow I gave above?

[madmanchan]

They are generally not documented anywhere. The current Camera Raw / Lightroom method generally preserves detail better on an upres, compared to the PS bicubic or bicubic smoother. However, on really big blowups you may prefer to use bicubic smoother since it is less likely to produce edge artifacts/chunkiness.

[bjanes]

Yes, LR's resampling algorithm is different from Photoshop's, with different strengths and weaknesses.

Eric,

I understand that ACR and probably Lightroom use the Lanczos algorithm (Jeff Schewe) whereas Photoshop uses a bicubic algorithm. I'm nor sure what this means in terms of practical results. I have not used Genuine Fractals, but understand that most experts state that it has no advantage is upsizing. Roger Clark has reported that one can double the effective linear resolution of an image with use of the Richardson-Lucy deconvolution algorithm. SmartSharpen in Photoshop is another deconvolution algorithm. Has it been improved in the current version of Photoshop and can it be used in this way?

Bill

[jjj]

I had a play around with enlarging images a while back and tested various methods. Genuine Fractals, using PS and using ACR to enlarge the RAW file, which is I guess the same as using LR.
ACR did by far the best job. Surprisingly good. The file used was a 20D file [chosen for it's content] and the bigger version looks better!
Notice the lack of jaggies on diagonal, the better sharpness on knurled part of the dart and sharper detail on the target.

 I posted some shots showing how useful it can be in this thread here.
JPEG look/ACR Calibration thread

[madmanchan]

Bill, you are correct in identifying which algorithms are used by which software. I do not actually know if Smart Sharpen has changed between CS3 and CS4 (sorry, I don't work in that area of the code). Nor have I attempted to use its methods for the purposes of image resampling (e.g., for large-scale upsizing).

Theoretically all "linear" filters, such as the cubic, Lanczos, etc. filters commonly used for resampling (including all the variants offered by software such as ImageMagick) will tradeoff in the areas of blurring, aliasing, and ringing. Blurring is obvious: if a filter blurs too much then you lose detail. Aliasing can generally manifest itself in terms of unwanted jaggies along edges, as well as false patterns (e.g., Moire patterns in fabric or window screens). Ringing generally shows itself as the presence of multiple dark or light halos surrounding a high contrast edge. It is theoretically impossible to eliminate all 3 with a single linear filter of this design. That doesn't mean you can't get good visual results, however.

More sophisticated filters can be used for the purposes of image resampling. They can in some cases produce superior results for large blowups. In general, however, they can in some cases produce unnatural (err, non-photographic) artifacts at the edges, and it can be hard to predict on an image-to-image basis whether you'll get one of these or not. (One "advantage" of a linear filter such as the ones described in the preceding paragraph is that they're so simple that it guarantees consistency from image to image ... e.g., if you get halos on one kind of edge, then you'll always see halos on that same kind of edge.) The other downsides of the more sophisticated filters are that they may not parallelize well, hence tend to take much longer to run (anywhere from 2x to 20x as long), and generally only make a difference for big upsamples.

In our view, most photographers tend to do image resampling using scales in the linear dimension of about 0.2 to 2. A scale of 0.2 is typically used when resizing an image down to web-sized display. A scale of 2 would be, for example, taking an uncropped 5D image and trying to make a 16" x 24" print at 360 ppi.

[Gandalf]

Unfortunately I can't post screen shots, but I recently did a 30"x45" print from a Nikon D2x. The difference between specifying the pixel size in LR vs 110% jumps in Photoshop with bicubic smoother was striking. I would say that Photoshop may have retained better detail, but the LR output eliminated the noise, making the detail more clear. After export of the Lightroom image (fully upressed to necessary print resolution) to Photoshop, I ran PKSharpener for capture sharpen, edge sharpen 1, and proper output sharpening for the device and resolution. The results speak for themselves.

I'm really not a big Lightroom fan, but I couldn't deny the results.

[Schewe]

The difference between specifying the pixel size in LR vs 110% jumps in Photoshop with bicubic smoother was striking.

What you did by doing step interpolation with Bicubic Smoother was basically ruin the image. Bicubic Smother was designed to eliminate the requirement to do step interpolation because that will "over-soften" your upsampe. So, you were basically misusing the tool and comparing THAT result to Lightroom was pretty much a DOH, what would you expect sort of moment.

Redo the test using Bicubic Smoother with a single upsample and then apply a dose of sharpening after the fact. Bicubic Smoother will do a better job of upsampling than Lightroom in that test–as it should if you are doing the right thing. Don't be surprised to see the Lightroom/Camera Raw resampling get "improved" as some point. Downsampling in particular can result in some ringing artifacts on certain images with certain frequencies and LR/CR upsampling tends to be less "sharpenable".

If you need to upsample, I suggest you read this...The Art of the Up-res

[httivals]

What you did by doing step interpolation with Bicubic Smoother was basically ruin the image. Bicubic Smother was designed to eliminate the requirement to do step interpolation because that will "over-soften" your upsampe. So, you were basically misusing the tool and comparing THAT result to Lightroom was pretty much a DOH, what would you expect sort of moment.

Redo the test using Bicubic Smoother with a single upsample and then apply a dose of sharpening after the fact. Bicubic Smoother will do a better job of upsampling than Lightroom in that test–as it should if you are doing the right thing. Don't be surprised to see the Lightroom/Camera Raw resampling get "improved" as some point. Downsampling in particular can result in some ringing artifacts on certain images with certain frequencies and LR/CR upsampling tends to be less "sharpenable".

If you need to upsample, I suggest you read this...The Art of the Up-res

So, Jeff, are you saying that uprezzing in Photoshop using bicubic smoother should give better results for say, making an 18" x 24" print from a Canon 5D at 360 dpi, than uprezzing/enlargening from Lightroom 2?  BTW, I read your article, and found it extremely helpful when I settled on my current methodology (about a year ago).  It's just reading that Lightroom may do a better job of uprezzing than Photoshop CS (as stated in my original post) that caused me to think maybe I should try that option and compare.

[Schewe]

It's just reading that Lightroom may do a better job of uprezzing than Photoshop CS (as stated in my original post) that caused me to think maybe I should try that option and compare.

Chris is wrong...

[mistybreeze]

Chris is wrong...

I remember a PhotoExpo seminar when Schewe told a frustrated photographer in his audience that "people who teach Photoshop are required to be certified." As if to say, just because you're Adobe certified, you speak truth to knowledge and possess the smartest information for professional users.

Well, we know how difficult it is to get two doctors to agree on anything. Why would we expect it to be any different with Photoshop gurus? Yet, there Chris Orwig stood, center-stage at the Adobe booth at PhotoExpo NYC, "teaching." And Schewe was nowhere to be found.

I missed Jeff Schewe at this year's PhotoExpo. Given his recent Camera Raw book, I was expecting (and looking forward to) a seminar on the newest version of Camera Raw. Instead, Adobe offered Orwig and Kolby. Honestly, what's a professional supposed to do?

[Schewe]

Well, we know how difficult it is to get two doctors to agree on anything. Why would we expect it to be any different with Photoshop gurus? Yet, there Chris Orwig stood, center-stage at the Adobe booth at PhotoExpo NYC, "teaching." And Schewe was nowhere to be found.

I have two books due by the end of the year...including the update to Real World Camera Raw due in a week or so and should be out in Dec. So, rather than go to yet another Photo Expo, I decided to stay here in Chicago to get work done.

Chris Orwig is a good guy...relatively new and perhaps without full knowledge regarding Photoshop's 3 flavors of Bicubic resampling and Camera Raw's Lancoz interpolation. Perhaps he has not done the extensive testing I've done (I'm pretty sure he wasn't involved with the original round of engineering that led to Chris Cox doing the Smoother/Sharper flavors of Bicubic, I know he's not directly involved in the Camera Raw development).

I actually wrote a section in the RWCR book last time (see page 139) that pretty much outlines the differences between Camera Raw and Photoshop's upsampling. If you read it, you'll see that Camera Raw's Lancoz resampling introduces ringing artifacts that make post upsample sharpening a problem. I'm also pretty darn sure (but can't really say anything due to NDA) that the Camera Raw engineers are aware of the issue and intend to do something about it at some point. Since that would change both Camera Raw and Lightroom's interpolation scheme, any opinions regarding Camera Raw's vs Photoshop's scheme would then, of course, change.

Honestly, what's a professional supposed to do?

If something is important to you, you should run tests, evaluate the results and make up your own mind. Letting others make your mind up for you (based on faulty data) is pretty much an abdication of one's professional responsibilities.

"people who teach Photoshop are required to be certified."

Pretty sure that's a totally mangled quote...pretty sure I said something to the effect that people who teach Photoshop SHOULD BE REQUIRED to be certified...and being an Adobe ACE expert means nothing.

[mistybreeze]

I have two books due by the end of the year...including the update to Real World Camera Raw due in a week or so and should be out in Dec. So, rather than go to yet another Photo Expo, I decided to stay here in Chicago to get work done.

I'm happy to hear you're busy writing. That talent does require a bit of quiet and concentration. Your absence at this year's Photo Expo made the event quite the dud. The most exciting booth: RayFlash. Even Tupperware salesman, Gary Fong, fared better than most.

Chris Orwig is a good guy...relatively new and perhaps without full knowledge regarding Photoshop's 3 flavors of Bicubic resampling and Camera Raw's Lancoz interpolation. Perhaps he has not done the extensive testing I've done (I'm pretty sure he wasn't involved with the original round of engineering that led to Chris Cox doing the Smoother/Sharper flavors of Bicubic, I know he's not directly involved in the Camera Raw development).

The nice thing about Chris Orwig: his photography style is very accessible to other photographers. Even if we don't shoot quite the same things or shoot them the same way, many of us can see an aspect of our work in his. So, when he uses his images to demonstrate Photoshop, it seems easier to learn because we can envision ourselves applying the same techniques. When that connection between teacher and student is made, it's easy to fall prey to clinging to every declaration.

If something is important to you, you should run tests, evaluate the results and make up your own mind. Letting others make your mind up for you (based on faulty data) is pretty much an abdication of one's professional responsibilities.

Ah, yes, time. Family or work? Friends or work? Lovers or work? Abdication produces costly pharmaceutical bills. I agree, one does have to sift and cull, and do deeper research to find the most beneficial truths.

Pretty sure that's a totally mangled quote...pretty sure I said something to the effect that people who teach Photoshop SHOULD BE REQUIRED to be certified...and being an Adobe ACE expert means nothing.

The quote wasn't verbatim but, I can assure you, there was no mention of "being an Adobe ACE expert means nothing," since there were plenty of ACE experts in the audience who paid to see you. You're not that reckless, are you?

Rumor has it that the Epson Print Academy tour is starting up soon. That certainly can make up plenty for your Photo Expo absence. But the link at the top of PhotoshopNews appears old and there's no mention of a new tour. May I suggest you take a moment to update the link?

[Hening Bettermann]

 @ jjj post #6
"I had a play around with enlarging images a while back and tested various methods. Genuine Fractals, using PS and using ACR to enlarge the RAW file, which is I guess the same as using LR."

jjj Your result looks indeed good, but how did you do that, doubling the size in ACR? I can find no such menu command, or option in the Save dialog.

--I think I found it, it's the Worklow Options reached via the link at the bottom of the image. Sorry!