Sharpening and Noise Reduction Sequence

[Brownbear]

Hi Colin,

I apologize in advance if I have overlooked something but when you speak of sharpening in these three steps are you using general sharpening or an unsharp mask? Or does your sharpening process work for both. Could you please clarify this for me?

Also, with respect to

(1) Avoid under-exposure (especially at high ISOs because the dynamic range of the capture decreases with increasing ISO, so the "safety margin" of a high-iso capture is significantly reduced to start with, resulting in more noise being made visible when the levels of the image are adjusted in post-processing - or in other words, if the image is exposed correctly in the first place - no raising of levels will be required - and the information captured will be as far away from the noise floor as possible.

I was trying to capture some photos of some eagles in flight in poor light (traveled far to do so, and tried even though it was impossible). I lowered my shutter speed to 1600, then 1250, Aperture 6.7 (as wide open as I can go with an extender on). I couldn't manage to get any closer. At an iso of 6400 the images were still underexposed, and yes they are full of noise.... If I had increased the iso further, to super high levels to increase the exposure, do you think the images would have had less noise even at higher iso's?

Thank you.

Hi Frank,

The "short answer" is "yes", apply noise reduction before sharpening, so as to avoid sharpening the noise. Unfortunately, the "long answer" isn't quite that cut and dried ...

In essence, noise reduction and sharpening are - to a very significant degree - "mutually exclusive" in the real world; sharpening increases the contrast around edges whereas noise reduction "averages" areas of a photo so that the impact of the (psudo-random) noise is reduced. It all sounds nicely "compartmentalised" in theory, but in practice, noise reduction ALWAYS ends up softening an image (so more sharpening is usually required to offset that softening). So in a very real way, in just the same vein as "sharpening will also sharpen noise" (and thus making it more visibly obvious), "noise reduction will also significantly reduce the effect of a given sharpening workflow", and thus make that more visibly obvious too.

So the question really becomes "which degrades an image more; sharpened noise or a soft image"? In real-world terms, the "A" answer is "avoid creating noise in the first place" - and in my opinion, that's very easy to do - AT ANY ISO. Simply follow 2 rules:

(1) Avoid under-exposure (especially at high ISOs because the dynamic range of the capture decreases with increasing ISO, so the "safety margin" of a high-iso capture is significantly reduced to start with, resulting in more noise being made visible when the levels of the image are adjusted in post-processing - or in other words, if the image is exposed correctly in the first place - no raising of levels will be required - and the information captured will be as far away from the noise floor as possible.

(2) Avoid excessively cropping an image. Noise is very small - far too small for our eyes to resolve at normal image sizes. Put another way; "if you don't crop an image excessively, the ONLY time you'll be able to see the noise in a correctly-exposed image is when viewing it at near 100% in photoshop".

Personally, I NEVER use noise reduction on my images -- I've just never found a need for it. If folks have problem noise then 999 times out of 1000 it's because they're either under-exposing the captures and revealing noise when compensating in post-processing, or they're cropping excessively -- both are issues that they should be fixing at time of capture, not during PP

On to sharpening ...

Sharpening is mis-understood by many -- a common "myth" is that sharpening only needs to be applied once (usually at the end of the workflow). Unfortunately, sharpening is needed to counter softening due to a variety of reasons - and because these situations occur for DIFFERENT reasons, DIFFERENT sharpening protocols are needed to counter them. So lets work through them with some real-world examples ...

CAPTURE SHARPENING

When an image is captured, a degree of softness is introduced due to 3 things (1) the digitisation process (the process of converting a continuous scene into individual pixels) (2) the anti-aliasing ("blurring") filter fitted over the sensor by the manufacturer to prevent weird patterns appearing when the sampling interval of elements of a captured image approach multiples of the pixel spacing on the sensor, and (3) the demosaicing process as the captured information is "re-constructed" during processing.

Capture sharpening is only visible at a pixel level -- so one has to zoom in to 100% to be able to see it. In the grand scheme of things it won't make any difference to the final image if it's not done ... it just makes the image nicer to work on at high magnifications when one is doing professional retouching. Time for some examples ...

Take the following image - straight out of the camera, with no sharpening ... (unfortunately I've had to down-sample it for display here, but I've tried to convey the effect none-the-less).



And now - since capture sharpening needs to be viewed at 100% magnification, lets grab a portion of the image (100% crop) - perhaps the portion I'd see when working at 100% magnification on my monitor (ie when 1 pixel of the image uses 1 pixel on my monitor) ...



You might think it looks OK (if you're not in the habit of applying capture sharpening then this will no doubt look pretty normal), but it's not optimal. Take a look at the same 100% crop below, but with capture sharpening of 300% @ 0.3 pixel appplied ...



If you look closely at the above 2 examples, you'll see that the 2nd example is "clearer" or for want of a better word "sharper" (doh!). As I say -- it doesn't make any difference unless you're looking at it at 100%

CONTENT / CREATIVE SHARPENING

I mentioned above that capture sharpening is only visible when looking at the image at 100% magnification ... when we look at the WHOLE image at once on our monitors, it may surprise some people when I say we're not actually looking at the whole image ... for me, a typical image from my camera is around 3600 pixels wide by 5600 pixels high -- but my monitor can only display about 1900 pixels wide by 1080 pixels high - so when I'm looking at an image (especially a vertical one) - I'm only seeing a small fraction of the actual information contained in the image -- and thus it's not hard to see how capture sharpening (which is typically done at a radius of 0.3 pixel) just can't be seen. Take a look at the image below - it's the same as the very first image, but it's had capture sharpening applied ... you won't be able to see any real difference ...



So - to make our images "pop", we need to apply more sharpening, but it needs to be sharpening that applies to bigger areas so that we can see it when looking at the full resolution image, but displaying this FULL RESOLUTION image at a smaller size so that it fits on our screens. So take a look at the image below ... I've applied content / creative sharpening of 75% @ 6 pixels (I'd normally use something closer to 40% @ 4 pixels, but I wanted the difference to be more obvious).



In real-world terms, content / creative sharpening should reveal the same additional "clarity" to an image the capture sharpening does, but with the image displayed at a size that allows the entire image to be viewed (not just a small portion of it, as is the case with capture sharpening).

OUTPUT SHARPENING

OK - so we've applied our capture sharpening - and we've applied our content / creative sharpening - but at this point we still have a full-resolution image - and full-resolution images aren't suitable for posting online because they're just too big. To get around this, we down-sample them to make them smaller. The algorithm that's recommended for doing this in Photoshop is called "Bicubic Sharper" (and it does a pretty good job), but I've found a way that's better - I use plain old "bicubic", but add a touch of sharpening afterwards (typically 50 to 100% @ 0.3 pixel again). The result is subtle, but if you look closely, you'll be able to see the difference ...



and "viola" - we're done!

Hope this helps

[rpcrowe]

I sharpen as both my first (after completing Camera RAW adjustments) and last procedure.

I begin post processing using Camera RAW and adjust the color balance, brightness, contrast, vibrance, etc. I do as much adjusting in Camera Raw as possible. Next, I use NIK Software to post process most of my images. I begin with the Dfine module to reduce noise and then apply the Sharpener Pro, Raw Pre-Sharpener. I don't have any sharpening dialed in my camera or Camera RAW. I then work with the Viveza module to adjust the image both globally and using the control points, to adjust specific areas. Sometimes, I use the Color Efex Pro module in the post processing mix, but, certainly not always. I will shift from Photoshop to NIK and then to Perfect Photo and back again, according to the needs of the image. I enjoy having the NIK and Perfect Photo as plug-ins rather than as stand-alone applications because that makes shifting between the three programs very simple. Each of the programs has its strengths and weaknesses.

When all of the above post processing is done, and before flattening, cropping and final sharpening, I will save the file as a Master Photoshop Document PSD file. That way I can crop this master any way that depending on the final product I want, flatten the layers and apply final Output Sharpening using the NIK Output Sharpener module; depending on that final use of the image.

The below workflow doesn't take pre-sharpening sharpening into consideration.



I first learned the master file concept when working exclusively with Photoshop but, have found that it meshes quite nicely when I use my NIK software and Perfect Photo Suite as Photoshop Plug-ins. BTW: I will use the onOne Perfect Photo Suite, especially for resizing; since a module of that Suite is Perfect Resize which was once called Perfect Fractals and was the standard of the industry.

[Colin Southern]

I apologize in advance if I have overlooked something but when you speak of sharpening in these three steps are you using general sharpening or an unsharp mask? Or does your sharpening process work for both. Could you please clarify this for me?

Hi Christina,

Most of the time I just use unsharp masks; they do the job nicely. on occasions where - say - a photo is a little out of focus and sharpening needs to be a bit more agressive to fix that - but it has high-frequency detail (like fine hair) that goes too crunchy, I'll just roll back the crunchy stuff using the history brush (or duplicate the layer - sharpen - and then erase (partially or fully) the crunchy stuff with a soft brush).

Also, with respect to

(1) Avoid under-exposure (especially at high ISOs because the dynamic range of the capture decreases with increasing ISO, so the "safety margin" of a high-iso capture is significantly reduced to start with, resulting in more noise being made visible when the levels of the image are adjusted in post-processing - or in other words, if the image is exposed correctly in the first place - no raising of levels will be required - and the information captured will be as far away from the noise floor as possible.

I was trying to capture some photos of some eagles in flight in poor light (traveled far to do so, and tried even though it was impossible). I lowered my shutter speed to 1600, then 1250, Aperture 6.7 (as wide open as I can go with an extender on). I couldn't manage to get any closer. At an iso of 6400 the images were still underexposed, and yes they are full of noise.... If I had increased the iso further, to super high levels to increase the exposure, do you think the images would have had less noise even at higher iso's?

Thank you.

It depends on what metering mode you're using and how the camera was setup.

Lowering the shutterspeed doesn't necessarily change the exposure because most automatic metering modes will change something else to compensate, so it can be a can of worms. For BIF you've really only got 2 variable in the exposure triangle you can play with (it's a given that your aperture will be wide open) - so if you're lowering your shutterspeed the camera will probably also lower the ISO, keeping the relationship the same, whereas what you're needing is exposure compensation so it it ups the exposure and gets the data further away from the noise floor.

Can you tell me more about how your camera was setup (eg make / model / shooting mode / other automatic settings that affect exposure?).

[Ken MT]

Colin, I am a little confused between the Unsharp mask which has Amount, Radius, and threshold, and the Adjust sharpening which has Amount, Radius and remove Gaussianblur, lens blur and motion blur. Which should I be using?

[Colin Southern]

Colin, I am a little confused between the Unsharp mask which has Amount, Radius, and threshold, and the Adjust sharpening which has Amount, Radius and remove Gaussianblur, lens blur and motion blur. Which should I be using?

Hi Ken,

I only ever use the USM. I'm sure that the smart sharpen is fine too, but for me, it's not worth the effort to tweak it.

Someone once said that "images are never really finished -- they're just abandoned by their creators at some point", and I kinda feel the same way. For me a USM is quick and easy and gets the job done so I can move on.

[DanK]

I'm not sure whether this will simplify matters or complicate them.

Like a number of other people here, I do most of my postprocessing in Lightroom. Colin pointed out that this is not optimal (at least for Canon raw files), but it works very well for most of my images. So, IF one does this, the situation becomes somewhat different.

Colin noted about capture sharpening:

Capture sharpening is only visible at a pixel level -- so one has to zoom in to 100% to be able to see it. In the grand scheme of things it won't make any difference to the final image if it's not done ... it just makes the image nicer to work on at high magnifications when one is doing professional retouching.

It makes even less difference if one is editing in LR or ACR, because AFAIK, the order in which one makes edits in these parametric editors has no impact on the final output. Therefore, I usually leave the "capture" sharpening at its defaults (although I have sometimes had the default set to zero) and sharpen at or near the end of the editing. If I am right about the way in which these editors work, the distinction between capture and creative sharpening is moot.

LR simplifies this in another way as well: it has output sharpening functions. They provide far less control than the method that Colin describes. LR gives you only a few options for the amount of sharpening, but if you find one of those options acceptable, it does the calculations for the size of the output. This is one of the few areas in which I have willingly (so far, at least) given up control over processing the image. I do all of my printing from LR, and I almost always create images for the web from LR also. I have found that the output sharpening options have worked fine.

So, in the ideal (simplest case), where I don't end up editing in Photoshop, my workflow for sharpening is very simple: sharpen to taste at 100% at or near the end of editing. Reduce the magnification to judge its appearance. (It is often helpful to look at the image at 50% before going smaller). When I am satisfied, print or output for the web using LR's output sharpening. For printing on my default luster paper, I use standard output sharpening, but for some other media, that doesn't work well. For metal prints, for example, I recently used the high level of output sharpening.

All right, I'm ready for the tomatoes.

[Brownbear]

Hi Colin,

Thank you for your reply.

1. Good to know about unsharp masks... However I am still confused because in ACR (capture sharpening?) the minimum radius is .5 and the max sharpening is 150, so when I tried to apply 300 and .3 to an image it wasn't possible. (Lightroom is similar). And to someone who usually uses .5 and .25 in raw, this seems extraordinarily high. ie; I tried to follow your suggestion on a try-out photo but couldn't do so.

2.

Nikon 7100 with a 300 mm Nikon lens (fixed aperture of 4) and for the eagles I needed an extender which was a 1.4.

Manual Mode Matrix Metering SS 1250 and an Exposure Compensation +1.67 to +2 ISO 6400

I began with an iso of 800, switched to 1600, then switched to auto iso with the max set at 3200 but I could see that the images were still underexposed, so I just set the iso to 6400 and they were still underexposed. (earlier in the day I managed a few shots with an iso of 3200 but they are just okay)

I think I recall reading in my manual that exposure compensation works best with spot metering (albeit the manual doesn't explain why). And upping the exposure compensation clips the white head and tail of a bald eagle.

And if I tried spot metering off of the brown wings of the eagle, wouldn't this also overexpose the white head and tail?

I am going to return on a sunny day, and try again but I really would like to know if there was anything else I should've tried. I think dropping below a SS of 1250 would make for blurry eagle photos (ie; in flight or an action pose)

Thank you.

Hi Christina,

Most of the time I just use unsharp masks; they do the job nicely. on occasions where - say - a photo is a little out of focus and sharpening needs to be a bit more agressive to fix that - but it has high-frequency detail (like fine hair) that goes too crunchy, I'll just roll back the crunchy stuff using the history brush (or duplicate the layer - sharpen - and then erase (partially or fully) the crunchy stuff with a soft brush).



It depends on what metering mode you're using and how the camera was setup.

Lowering the shutterspeed doesn't necessarily change the exposure because most automatic metering modes will change something else to compensate, so it can be a can of worms. For BIF you've really only got 2 variable in the exposure triangle you can play with (it's a given that your aperture will be wide open) - so if you're lowering your shutterspeed the camera will probably also lower the ISO, keeping the relationship the same, whereas what you're needing is exposure compensation so it it ups the exposure and gets the data further away from the noise floor.

Can you tell me more about how your camera was setup (eg make / model / shooting mode / other automatic settings that affect exposure?).

[Richard Lundberg]

i think someone else mentioned Focus Magic in the thread. If you are otherwise happy with the image, either USM or Smart Sharpen will crispen, but I use Focus Magic before sharpening if there's any question about focus or movement.

[Colin Southern]

1. Good to know about unsharp masks... However I am still confused because in ACR (capture sharpening?) the minimum radius is .5 and the max sharpening is 150, so when I tried to apply 300 and .3 to an image it wasn't possible. (Lightroom is similar). And to someone who usually uses .5 and .25 in raw, this seems extraordinarily high. ie; I tried to follow your suggestion on a try-out photo but couldn't do so.

The general rule is "anything you CAN do in ACR you SHOULD do in ACR", but for me anyway, capture sharpening is the exception; I do it in Photoshop for this very reason. There's probably an equivalent in ACR, but I've never bothered to research it -- for me it was easiest to just stick with Canon's recommendation (300% @ 0.3 with threshold depending on ISO due to noise) - and for that, in Photoshop, it's as easy as one click of the mouse. Additionally, I've also found that if I zap dust bunnies before capture sharpening than it makes the job a little easier.

Nikon 7100 with a 300 mm Nikon lens (fixed aperture of 4) and for the eagles I needed an extender which was a 1.4.

Manual Mode Matrix Metering SS 1250 and an Exposure Compensation +1.67 to +2 ISO 6400

I began with an iso of 800, switched to 1600, then switched to auto iso with the max set at 3200 but I could see that the images were still underexposed, so I just set the iso to 6400 and they were still underexposed. (earlier in the day I managed a few shots with an iso of 3200 but they are just okay)

I think I recall reading in my manual that exposure compensation works best with spot metering (albeit the manual doesn't explain why). And upping the exposure compensation clips the white head and tail of a bald eagle.

And if I tried spot metering off of the brown wings of the eagle, wouldn't this also overexpose the white head and tail?

I am going to return on a sunny day, and try again but I really would like to know if there was anything else I should've tried. I think dropping below a SS of 1250 would make for blurry eagle photos (ie; in flight or an action pose)

Thank you.

I'd be inclined to use Av mode and then see what shutterspeeds the camera offers at various ISOs. At the end of the day, you have aperture maxed out, so only SS and ISO to play with ... and you need the SS to be at a certain level to freeze motion - so if it were me, I'd whack on something like +1 EC Av mode and increase the ISO until I had the best compromise of noise & motion freeze.

I'd also just stick with matrix metering; the sky is going to be by far the most stable reference for exposure. If you're shooting with the light behind you it should be fine; the worst combination is shooting a dark bird into the light.

[Brownbear]

Thank you Colin. Truly appreciated.

Happy New Year!