Capture sharpening/ Output Sharpening
rutt
Registered Users Posts: 6,511 Major grins
I've been thinking about the idea of separate input and output sharpening which has won a lot of converts since I learned how to sharpen from Dan Margulis and wrote some dgrin tutorials on the topic, here and here (there might even be a few more about HiRaLoAm sharpening somewhere.)
I think I understand the theory behind input/output sharpening: Sharpening exists to overcome aliasing problems caused by the pixel grids of the camera sensor and the printer or monitor. The resolution of the sensor grid is very different from the resolution of the printer (say) ink dots. What's more size of the printer's dots relative to the image itself will vary depending on how large we want to print the image.
So, it seems natural to sharpen once in the very beginning of the workflow (input sharpening) to correct for the camera sensor and again at the very end of the workflow with knowledge of size of the final print and specific printer to correct for the printer.
This is roughly the idea, right?
I've sort of an old dog, and don't seem to be able to learn this trick. I still sharpen once (well twice with HiRaLoAm) at the very end of the workflow to make the image look good on my monitor when viewed at 100% or 50% or 25% magnification (whatever fits and shows what I consider the most important elements of the image.) I think my results look very good and so does everybody who sees them.
I tried to adapt the input sharpening idea, but I found that it often led to bad things down stream in my workflow, most often visible sharpening artifacts. And yes, I tried using LR sharpening with its mask. Sometimes it wasn't halos exactly, but harsh edges. I guess I might behoove me to try again and show my results if I can reproduce these problems.
I haven't gotten my head around later versions of LR which offer output sharpening. Maybe I have to do that before really making my point? Or does it apply to a PS centric workflow as well?
I'd really like to get beyond the theory on this and see real world examples from people who have adapted input/output sharpening and feel it has improved their results. If there is something to learn, I'd like to learn it.
No pure theory, please, except to correct my statement of input/output sharpening theory, which I tried to present impartially.
I think I understand the theory behind input/output sharpening: Sharpening exists to overcome aliasing problems caused by the pixel grids of the camera sensor and the printer or monitor. The resolution of the sensor grid is very different from the resolution of the printer (say) ink dots. What's more size of the printer's dots relative to the image itself will vary depending on how large we want to print the image.
So, it seems natural to sharpen once in the very beginning of the workflow (input sharpening) to correct for the camera sensor and again at the very end of the workflow with knowledge of size of the final print and specific printer to correct for the printer.
This is roughly the idea, right?
I've sort of an old dog, and don't seem to be able to learn this trick. I still sharpen once (well twice with HiRaLoAm) at the very end of the workflow to make the image look good on my monitor when viewed at 100% or 50% or 25% magnification (whatever fits and shows what I consider the most important elements of the image.) I think my results look very good and so does everybody who sees them.
I tried to adapt the input sharpening idea, but I found that it often led to bad things down stream in my workflow, most often visible sharpening artifacts. And yes, I tried using LR sharpening with its mask. Sometimes it wasn't halos exactly, but harsh edges. I guess I might behoove me to try again and show my results if I can reproduce these problems.
I haven't gotten my head around later versions of LR which offer output sharpening. Maybe I have to do that before really making my point? Or does it apply to a PS centric workflow as well?
I'd really like to get beyond the theory on this and see real world examples from people who have adapted input/output sharpening and feel it has improved their results. If there is something to learn, I'd like to learn it.
No pure theory, please, except to correct my statement of input/output sharpening theory, which I tried to present impartially.
If not now, when?
0
Comments
I do Capture Sharpening in LR or ACR, either one, along with correction for chromatic aberration ( I find that step very important even with good L glass and I do it at 300% or greater ) and then Local Contrast Enhancement ( rather like your HiraLOam I believe ) in Photoshop, and then finally Output Sharpening in the print module in Lightroom. The output sharpening is not done on the image file I have saved in Lightroom, only on the file immediately before it goes to the printer since image size is a factor in how LR does that sharpening. "Output Sharpening" is a black box, I have no idea what is actually being done, just that it seems to work for my images with my Epson 3800.
Is this the best way to sharpen, I actually do not know. What I do know, is that I have several dozen large prints, 16 x 20 or bigger, that, to my eye, are vastly sharper than I ever saw with darkroom prints, and that I cannot see artifacts degrading the image. You may have a more discerning eye than mine. I will be happy to do a print exchange again with you to see if you agree with my opinion, or not.
I work this way because I find it fast, easy, consistent, and "good enough" If I cannot see defects, I rather doubt most viewers will either. I am also aware that you are not Most Viewers, and I am eager to see if you can show me a better, faster, easier way of doing things.
I think I have an open mind, in so far as editing is concerned. Faster, easier, and better or at least "good enough" is always worthwhile.
I am sure Andrew Rodney will be along to give his opinions in this regard also.
Moderator of the Technique Forum and Finishing School on Dgrin
It may be that capture sharpening is incompatible with other parts of my (PS centric) workflow. Fine. That would be good to know.
Output sharpening is about anticipating the softening which will occur in the output process and pre-sharpening so that smoothing won't roll of the edges. Half-tone grids in printing, dot patterns on ink-jets and pixel bleed on LCD displays will all soften hard edges to a degree. Ouput sharpening deliberately exaggerates edges in the file so that when the image is printed/displayed the softening of the output device will bring them back to where they should be. Perfect output sharpening should be done differently for each different output device and display size.
Thanks. That's a very clear statement of the theory and it makes a lot of sense. I think I did understand it already, but couldn't have given the details as well as you did with all the correct terms.
I hate to say it but understanding the theory is the easy part. The question I'm asking is an empirical one: how much difference does this actually make compared to my perhaps less theoretically sound practice? And is that capture sharpening somehow incompatible with a somewhat PS & LAB centric workflow? Perhaps that's why I have trouble with it? (I know I should provide examples of this.)
Is there someone out there who thought s/he was good at sharpening before the capture/output paradigm came on the scene and then was converted by quality considerations (as opposed to Jim who seems to have defaulted into a LR centric workflow for convienience?) And if so, can that person explain what happened and give examples? Even better, can s/he explain what else had to change?
Moderator of the Technique Forum and Finishing School on Dgrin
Sharpening works by recognizing edges which have been blurred and enhancing them to make them sharp again. The more the edge has been blurred, the larger the radius is required for the sharpening to catch the edge and enhance it. However larger radius sharpening has artifacts: both false positives (enahncing things it shouldn't) and less that perfect edge shapes. Being careful to preserve your edges throughout your process means you can use a lighter touch on the output sharpening and which can create a better final image.
That said, if you are starting with a high quality 21mp RAW file and printing 4x6, capture sharpening won't make any difference because you will be down rezing before you print anyhow.
As for workflow, I find it best to do all my big moves: white balance, exposure, shadow and hightlight recovery, rough brightness and contrast as part of the RAW conversion and capture sharpening. I don't try to make it perfect; I just get it in the ballpark before I go to Photoshop for refinements. However if I use a default RAW conversion under the theory that I can fix it all in Photoshop, I find I lose some the of the crispness of the original image.
LiquidAir, this is where things get interesting!
Have you compared an unsharpened raw image downsized against a capture sharpened raw image downsized to the same size - with the same output sharpening applied to both images? Do you know of a web page that goes into this? If so, what were the input and output pixel dimensions and what was the output? Further, does this still hold true/false with different input/output dimensions etc? What resampling method was used?
What if one is not resizing/resampling down (averaging), but using the pixels as is - so only sharpening is being evaluated?
What if one is resizing/resampling up - inventing pixels?
As John said, he is not looking for "theory" - that is understood, he wants real examples or comments from those that know and can demonstrate the facts at hand.
Unless the output is monitor viewing, evaluations have to be performed on actual prints - not pixel peeping...
Does the emperor wear no clothes, or some clothes, or only wear clothes in some circumstances?
Stephen Marsh
http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/
http://members.ozemail.com.au/~binaryfx/
http://prepression.blogspot.com/
Sorry, I wasn't inventing anything that you didn't say yourself:
And I don't really think I can show you a faster, better way of doing things. It sounds like you are getting results that you like with relatively little work. And I know you are fussy.
I guess I really have three specific questions:
I don't do this. Have you written up something on it or can you point me somewhere?
It is on my list - I will try to get a "roundtoit" for you. I learned most of it from the videos on Luminous Landscape and watching Marc Muench, and just playing with the sliders a bit along the way. I am sure it is covered in the usual encyclopedic tomes about CS4, for those who are willing to actually read books these days. I say that knowing that you are a reader, John. I am trying to leave out my smilies for you also.
One of the things I have learned is that some excellent, highly regarded lenses do exhibit CA at times, and I have couple Tamron travel zooms that vary a lot, from some frames with no CA apparent, and others with significantly noticeable CA. The point being that its presence can not always be predictable, and it needs to be looked for at 300% in the corners of each image. Primes, that do exhibit CA, can usually be corrected with similar values of the red/green and blue/yellow sliders, but not always.
Chromatic aberration will be worsened by sharpening too.
You can correct for CA in PS, but I never did, while in ACR it is relatively straightforward to correct 80-90% of it. You cannot always correct it totally.
My ROUNDTOIT can be found here - http://www.dgrin.com/showthread.php?t=159524
Moderator of the Technique Forum and Finishing School on Dgrin
This is my pratical experience:
I shoot with 12MP Canon 5D (classic) and my printer is an Epson 3800. Generally I print at 360ppi because I can't see any improvment from higher resolutions. When printing a 4x6, the file I send to the printer is 3MP, so I have scaled my source file down by a factor of 2 (linear).
Lets say I my capture sharpening is 1 pixel radius and my output sharpening is 1 pixel radius. Because of the scaling for print, my effective output sharpening raduis is twice the size of my capture sharpening. Under circumstances like this I find that capture sharpening makes little if any difference in the quality of the output. However, when I make a 16x24 print from the same file, my output sharpening has a smaller radius than the capture sharpening so the capture sharpening makes a quite noticable difference in the final ouptut.
In practice my output sharpening radius (for print, not the screen) is usually a bit larger than 1 pixel and my capture sharpening radius can vary from about 0.7 pixels to about 2 pixels depending on the quality of the RAW file.
The primary reason to always apply capture sharpening is really workflow. I find it much easier to treat issues in the capture on my first pass through the image and then process for output when I have decided what my output format is going to be.
In practice I typically apply a light default capture sharpening to all my images in Lightroom and I only tweak that if the image has significant technical flaws. Then, if I decided I want to print large (bigger than 8x12) I go back and reprocess the image from the RAW paying careful attention to the capture sharpening stage.
Overall yes. We view images in a continuous tone world. Digital images are (mostly) composed of a series of square pixels. Look at a circle you make at differing resolutions in Photoshop at 100% and greater or lesser zoom ratio’s and you can see that we have to build such images using squares. The more of them, the harder it is to detect they are composed of a circle built of squares. On top of that, each capture device has differing degrees of “sharpness” and resolution. So the idea of one size fits all capture sharpening doesn’t wash. Then we have the disconnect between differing output devices and how they build dots to again, simulate a contone appearance. So the idea that you have some single (or even a few) settings that produce ideal sharpening is nonsense. Images have content that are different. There are high and low frequency images. For example, scan a “line” across any image from left to right. If this is an image of a face, by and large, there are lots of smooth, low frequency areas. There are few alternating lights and darks lined up in a row. Take a scene of a forest with many trees and branches. Scan the same single row and you see lots of alternating lights and darks; high frequency. Sharpening the same degree will either produce a portrait that is too “crunchy” on output or a forest that isn’t sharp enough. Good sharpening techniques can build a mask from the image itself to control what does or doesn’t get sharpened. You don’t want to sharpen the dark regions where all the noise lives versus the lighter tones. You don’t want to or need to sharpen smooth areas like the sky or the skin of a portrait. So again, the idea of using one set of values is silly.
Look at how advanced the masking is in Lightroom and ACR when you use Capture Sharpening and hold down the alt/option key and move the sliders. LR builds a mask on the fly where you can see what’s going on behind the scenes. Its light years ahead of using Photoshop, plus you are doing this non destructively and can alter the effects at any time.
Capture sharpening should be based on the full resolution of the original while output sharpening is based on both the output device, which will differ for many and the number of existing pixels for output. Same device, a 5x7 and a 30x40 would require vastly different output sharpening. But the capture sharpening is “fixed” based on the original, and the output sharpening after sizing using this original capture sharpening. Like capturing in a wide gamut color space in high bit, this capture sharpening is applied to the master archive upon which you can subsequently output sharpening to any device and to any size when it is time to print the image. As Bruce states in his original article, it is very much like sound color management. Scan (capture) once, output many times, using data that is appropriate and correct for all subsequent work.
If you live in a Dan world (scan once, output once), the idea of scanning to size and sharpening to size is fine. But today, modern workflows are more flexible in terms of size, color space, output devices, and in this case sharpening. The idea of a minimum of two rounds of sharpening address this modern workflow!
Exactly!
Sharpening for a good looking display is great when the output is to a display. Its usually far from ideal when the output is something vastly different than a display (a very low resolution output device). Visually sharpening is simply not effective for multiple output devices because they differ so from a display.
But they looked OK on the display ;-). Yes, you see how this idea doesn’t work.
Capture sharpening should be subtle. And yes, you are supposed to work somewhat visually although I’ve found that once I have a setting that works well for my specific capture device (in this case a 5DMII), I just make a preset. Less sharpening is better than too much and yes, the LR team does allow you to go over board. Another problem with visual based sharpening is how if you go too far, then back off, the visual effect you see is the image getting softer and your brain plays a trick on you. I believe this is why we see so many users over sharpening their images. If they back off appropriately, they see the image look less sharp and over sharpen. Best to look away from the display or take a break and revisit the image after awhile.
It was introduced in version 2. You find it in the Print module. Its based on Inkjet output.
Author "Color Management for Photographers"
http://www.digitaldog.net/
If I had a smidge of your knowledge on the subject of sharpening I would never get anything done :wow
Just messing with you guys a little bit...actually I am jealous of your level of knowledge on the subject.
I am going to school on this thread...
http://www.flickr.com/photos/21695902@N06/
http://500px.com/Shockey
alloutdoor.smugmug.com
http://aoboudoirboise.smugmug.com/
Given that, under many circumstances, the lens is the limiting factor for resolving power and sharpness, it is hard to separate lens corrections from capture sharpeneing. When printing large I often use DxO for my RAW conversions because it provides a complete, integrated package of conversion, lens corrections and sharpening. I can get the same (or even better results) using other tools, but is much more work to do it by hand and often DxO is good enough.
Actually, I don't think I made myself 100% clear. Without capture/output sharpening, I get what I think are great looking results, both on the monitor and in print at all sizes. But, I found that capture sharpening ends up making my image look bad later in the workflow on the display, in print, everywhere. I just can't make it work with the rest of my workflow, even before "output" sharpening.
Capture sharpening alone shouldn’t produce such results (its very subtle). Something else is going on here (maybe too much adjustment on the sliders or double sharpening). You are working with raw files right?
Author "Color Management for Photographers"
http://www.digitaldog.net/
I have problems combining ACR capture sharpening with a downstream PS workflow which ends with what you might call output sharpening. So, yes, there is double sharpening, but isn't that the idea of input/output sharpening? And for that matter, I get just about the same kinds of problems from sharpening lightly in PS at the beginning of my workflow (which is how I imagine that capture sharpening was first tried and tested.)
But I start not liking the results before that point.
One of my original questions was whether the rest of my workflow, PS centric, often using LAB, frankly mostly learned from Dan, really isn't compatible somehow with the capture/output sharpening idea, which as I said, sounds good in theory.
Andrew, I really like the results I get, so I'm not ready to completely toss the way I work. But I am always interested if there is an improvement I can make. So far, honesty, I haven't found it.
The two rounds of sharpening have to be used in tandem. That is, output sharpening in ACR or LR is based on its capture sharpening. Using output sharpening that isn’t part of this process with associated with capture sharpening isn’t going to fly either. You need to either turn off all capture and output sharpening in ACR/LR and do it in Photoshop or do both in the raw converter and forget using Photoshop (which I would submit is faster, easier and more flexible). So yes, two rounds of sharpening is the idea but two rounds that understand both processes if you will. If you want to test the ACR/LR sharpening workflow, based on Bruce’s work, don’t add any additional sharpening other than creative sharpening which is a subjective, artistic process that isn’t implemented in this raw workflow.
Author "Color Management for Photographers"
http://www.digitaldog.net/
I think the key here is that viewing the file at 100% on screen is not an output format. In my workflow, I apply capture sharpening and then do all my processing in Photoshop to create my reference, archival version of the image. Then, when I have decided on an output format be it an 8x10 print or an 800px screen display, I apply sharpening as part of the output process. However I don't burn that second sharpening pass back into my reference copy because I my decide to output that reference copy in some other format later.
Some caveats to this process: online services like Smugmug and Flickr sharpen as part of their scaling so, unless you intend the image to viewed at 100%, you shouldn't apply output sharpening before upload. Also, I believe (Andy can correct me on this) output sharpening is built into Smugmug's print process. In practice the only times I actually apply output sharpening are when I am printing at home (since I got my 3800 that's how I make most of my prints) and when pre-scaling for the web.