Skip to main content
Adobe Photoshop Family

4.5K Messages

 • 

76.3K Points

Wed, Apr 6, 2011 9:03 AM

9

Lightroom/Camera Raw: Improve Sharpen Masking

I love everything about the new sharpener, except the masking artifacts.

There is a tell-tale artifact of the masking that I call "sparkling", which I've grown to dislike so much I rarely use the masking anymore.

Its sort-of reminiscent of the masking halo artifact left by brushing - the sharpening spills too much out of the edgy area and into the smooth areas, leaving a sharpening sparkle around the edges of things.

Its unfortunate, since the concept of masking is great, and the UI for it is awesome (alt key to see where it will apply).

Some way to "pull in the sharpening effect" around the edges, so to speak, (or something like that), to eliminate sparkling would by greatly appreciated.

PS - Reducing detail helps, but then of course you lose detail in sharpened areas...

Rob

Responses

946 Messages

 • 

13.8K Points

10 years ago

I have masking set up in my defaults, and I've never noticed any "sparkling". Can you post some examples of similarly-sharp images both with the sparkle and without it? The "without" will obviously have to be sharpened using another method.

164 Messages

 • 

5.2K Points

10 years ago

I don't use auto masking any more for the exact same reason. It simply doesn't work well enough.

4.5K Messages

 • 

76.3K Points

I've heard people say things like "I use it all the time on photos I dont care about..." - meaning, its a nice convenience when speeding through a shoot, but if you care about quality - just say "no"...

4.5K Messages

 • 

76.3K Points

10 years ago

Settings chosen to exagerate the problem, so it looks pretty bad, but here's one:



Note: Crustiness around the edges I call sparkling.

Sharpness: 70
Radius: 1.5
Detail: 80
Masking: 80

Lum. NR: 17

4.5K Messages

 • 

76.3K Points

10 years ago

And here's one with same sharpen settings, but detail turned way down, and masking turned all the way off. I also reduced luminance noise reduction some.



Note: This was not a sharp picture to begin with, and was shot with a D300 @ISO3200.

Note: less crusty around the edge junctions.

Sharpness: 70
Radius: 1.5
Detail: 10
Masking: 0

Lum. NR: 9

4.5K Messages

 • 

76.3K Points

PS - Color NR detail was set to 80, which is too high for this photo. Reducing it to around 20-30 reduces the apparent CA and softens the hard edges some.

4.5K Messages

 • 

76.3K Points

10 years ago

Note: I realize this is not a "fair" comparison.

As I said before: one can reduce the sparkle by reducing detail, and the extra Lum. NR in the above shot tends to oversmooth the areas without the sparkle, which was the purpose. The point of this exercise was to demonstrate the phenomemon to which I am refering, which I think this example accomplishes.

PS - A good way to compare these two shots is to download the images, then switch back and forth..

R

248 Messages

 • 

4.1K Points

Rob - it looks to me like your image has CA (along the lower edge of the flower) that you are sharpening. Try removing the CA and see what happens.

4.5K Messages

 • 

76.3K Points

This was shot with a Nikon 70-300 at 300mm/5.6(wide-open) - read: CA is generally bad with this lens at these settings. CA is as corrected as it gets in Lightroom. Note: the CA is absent in NX2 and DxO (I checked). Anyway, it doesn't change the phenomenon - look all over...

248 Messages

 • 

4.1K Points

Rob - do you get this problem with other lenses and at base ISO? Also, what does equivalent sharpening look like in NX2? I tried an image taken with a D700 and 105VR, ISO 800, f/8 and I just tried maxing out both sharpening amount and detail (100), with the radius set at 2. These are extreme settings and I do not see any sparkling halos. On one bright leaf there is a very small amount of CA that lightroom does not correct that is exacerbated by sharpening as you would expect.

4.5K Messages

 • 

76.3K Points

10 years ago

This phenomenon seems to be noticed by and bothers some individuals, but not others. If the shot has very little noise, or very little sharpening, then its barely noticeable. It becomes more glaring as noise and sharpening increase. The reason I chose a noisy unsharp photo was to make it obvious. Its not tied to the lens. It is tied to the masking. It could be remedied by an adjustment to the masking algorithm.

946 Messages

 • 

13.8K Points

Perhaps I don't usually see it because I virtually never crank up detail beyond 25. My defaults are 25 detail and 35 masking, and I use regular sharpening as needed.

4.5K Messages

 • 

76.3K Points

10 years ago

DxO:

4.5K Messages

 • 

76.3K Points

10 years ago

Surprised me how good this photo turned out in DxO. I purposely tried to process it to look the Lightroom version.

4.5K Messages

 • 

76.3K Points

10 years ago

NX2:

Note: I did not spend as much time working the NX2 image as I did the others, but should still give an idea...

513 Messages

 • 

11.1K Points

10 years ago

Rob, I don't think you should change the "Detail" settings between comparisons.

I wonder whether the crustiness you are seeing isn't just noisy sharpening halos, i.e., has nothing to do with the masking. In this case masking would only highlight the noisy sharpening halos because the rest looks smoother, but wouldn't introduce them.

I tried to replicate your problem and in my experiments masking never introduced crustiness but only made it more visible. I only tried two photos so I'm not saying you don't have a point, but it might be worth posting a comparison where only the masking value is changed and everything else is left constant.

4.5K Messages

 • 

76.3K Points

10 years ago

There's no question where it comes from: The mask encompasses edgy areas with a fading from edgy area to smooth area. That works out nicely in broad areas that have gradual transitions between edgy and smooth, but across hard edges the noise is being sharpened in areas that should be smooth bokeh. You can tell by looking at the mask that the algorithm has two rankings: edgy and smoothy, and the mask value just sets a threshold. It needs to be modified for hard edge to smooth bokeh transition detection and not spill the sharpening into the bokeh. Or something like that. I mean, other software does not have the same artifacts (see DxO & Nx2 above). Sometimes they have other artifacts instead, but my point still stands. I've commiserated about this in private with several other people who can't understand why this isn't noticed by every critical eye, and bothers all of them. I've brought this up a few times before in the forum, and always with the same generally cool reception, except for a couple of people saying "Amen". The purpose of the above examples was to illustrate the phenomena, not to compare different methods of sharpening. You can see from the top picture, that the detail sharpening extends too far beyond the edges of things. I mean maybe my under-the-hood speculation and/or proposed solution is laughable to those who understand the inner workings, but the phenomenon is offensive enough to me and some others that we simply dont use it (except as a quick solution for photos we dont care too much about...). If we are in a small enough minority that Adobe wont fix it, then so be it, but I hope something gets done about it.

946 Messages

 • 

13.8K Points

"There's no question where it comes from"

Right - from detail. Think about it. Masking something only reduces it, not enhances it. What you're demonstrating is that shutting off sharpening in most areas and cranking up detail to compensate isn't such a good idea. I agree.

Detail is for enhancing pixel-level detail that's already there. Since noise is really no different from pixel-level detail (to the application) it's going to enhance noise a lot so it's best used sparingly on high-ISO images. I've found that pushing detail above 50 is generally too much for ISO 100 images. I generally make heavy use of that type of sharpening only on stacked images that have virtually no residual noise.

My suggestion: Leave detail alone at the default, and move it at most 10 points in either direction. Keep masking in the range from 0 to 40 or so and use sharpening and radius in concert with those ranges. Next, on high-ISO images use L-NR as necessary but start with its detail setting at 75 and adjust between 60 and 100 as necessary.

I've sharpened thousands upon thousands of images with LR and I've never once seen the effect you have demonstrated here.

513 Messages

 • 

11.1K Points

I concur with Lee Jay's analysis. Your problem doesn't seem to be "too little masking" but "oversharpening" or at least "inappropriate sharpening".

If the mask for the smooth area further transitioned towards detail then in case there is true detail (as opposed to noise) some of that detail wouldn't appear with the same acuteness anymore.

Would it be appropriate to rephrase your problem as one having to do with noise reduction? Do you see the "sparkling" in images with very little noise as well? If not, I guess you are asking for noise reduction that still allows subsequent sharpening to extract satisfactory detail.

I'm not trying to deny that there might be a problem but before one changes anything it is good to identify where the problem is really coming from. You might be right after all, I'm just wondering.

4.5K Messages

 • 

76.3K Points

I would have no objection to a solution that integrates luminance noise reduction with sharpening. I mean, in most photos, I either want to reduce noise OR sharpen a particular region, but not both. In fact, I could even see a solution via using the exact same algorithm for masking as noise reduction, but allow a different threshold for the sharpen and noise reduction masks. (and of course the NR operates on the smooth areas, instead of the edgy areas). That way I could set the sharpen mask at say 40 to sharpen most edgy areas, but spilling into the bokeh more than I'd like, then set the noise reduction mask to say 60 which would NR all the smooth areas pushing further into the edgy areas to help eliminate the sparkle. Hmm... - I may be on to something here... Thinking more: It may make more sense to use the exact same mask, but have a bias setting to indicate the NR vs. sharpen weighting in the transition zones (effectively adjusting the mask to confine the sharpening...). This would also go further to eliminate the sparkle just because less sharpening would need be done to attempt countering the NR, and less NR would need to be done to counter the effects of sharpening...

513 Messages

 • 

11.1K Points

So it appears that "sparkling" is only a problem with noisy images. I see your point about confining sharpening and NR to a mask and its invert plus a way to define the transitioning between the respective areas.

However, I'm not experienced enough in image processing to judge whether that would really be a good idea. It seems to me that you first have to remove noise and then can start to sharpen. If you do the latter without the former, your sharpened, noisy areas will have lots of artificial detail and look quite different in character to the unsharpened, denoised parts.

I would be against treating NR and sharpening as opposite ends of a spectrum. It works for the clarity slider because it wouldn't make sense to both decrease and increase mid-tone contrast at the same time, but I feel it makes sense to both denoise and sharpen at the same time. I'm wondering whether the "sparkling" examples are just examples of trying to use masking as a replacement for prior denoising.

4.5K Messages

 • 

76.3K Points

One of the things I like about NX2 is that both noise reduction and sharpening can be be done entirely locally. And more often than not, I lay a global base coat of both NR & Sharpening, very minimally, then do NR & Sharpening locally, but never in the same region (maybe overlapping in some regions) - buffing out the bokeh with NR, and crisping up the edges and/or subjects with the sharpener. I also regularly sharpen different regions with different settings. Anyway, if Lr had the ability to apply NR locally, preferably backed with a good auto-masking technology, and sharpening with different settings, that may be enough to assuage...

513 Messages

 • 

11.1K Points

"Anyway, if Lr had the ability to apply NR locally, preferably backed with a good auto-masking technology, and sharpening with different settings, that may be enough to assuage...": I see you are already following the "adjustment brush for noise reduction" topic. I hope that if anything like this is realised then not with an additional slider on adjustment panels, but as part of a generalised "selection & adjustement" scheme in which you can define masks in various ways that combine (e.g., painting with brushes, polygons, mask gradients (<- graduated filter)) and then associate arbitrary adjustments with them.

4.5K Messages

 • 

76.3K Points

Yup.

4.5K Messages

 • 

76.3K Points

10 years ago

When I do use the sharpen masking, I generally set Detail on zero. Lightroom's PV2010 raw conversion algorithm is inherently very detailed, and as you've said: its hard to tell the difference between true detail and noise detail in the sharpening of it. I have presets based on ISO that assign a default detail between 25 (for ISO100), 20 (for ISO-200), ..., to zero for ISO3200... Its true that I never noticed this phenomenon for the first year I used Lightroom, but now I'm very sensitive to it. Anyway, yes: best to recover detail in high-ISO shots using the NR detail sliders if possible before resorting to sharpening detail.

Summary: masking is not creating the artifacts. the artifacts come from sharpening detail all around edge areas, and the sharpening, unfortunately extends too far into the smooth regions for my taste, when using the sharpen mask.

PS - Even at detail = 0, there is considerable "detail" sharpening...

I mean, maybe the solution is not in the masking, but in the sharpener itself - something that can separate more linear edges and leave the "sand" alone - I dunno. I've never written any image processing software in my life...

946 Messages

 • 

13.8K Points

"I mean, maybe the solution is not in the masking, but in the sharpener itself - something that can separate more linear edges and leave the "sand" alone - I dunno. "

There's always room for improvement in the smooth-but-steep-transitions to crunchiness ratio.

4.5K Messages

 • 

76.3K Points

10 years ago

Lightroom again:

Sharpening:
Amount 70
Radius 1.5
Detail: 10
Masking 65

Lum. NR: 17

i.e. same settings as original post, except detail left low, and masking reduced.

Intended for comparison purposes (as opposed to exaggerating the phenomenon like the first sample).

These are more reasonable settings, and the sparkling phenomenon is far subtler, but still existent.

I think a simple change to reduce detail sharpening in transition zone between masked and unmasked areas would do it. So full detail is rendered only in areas not near boundary between masked and unmasked regions, and zero detail is rendered in transition area bordering on masked region - blended in between. As it stands, full detail is being sharpened if its unmasked, and zero sharpening is being applied if its masked, and there is no graduation, no blending in the transition zone, making for an abrupt transition between masked and unmasked which is why the sparkling.

Note: this phenomenon also occurs in the bokeh - spurious detail sparkling occurs due to edge threshold being exceeded a smidge - little patches of unmasked areas with sharpened detail dance with the adjacent completely masked & smooth bokeh. The above mentioned change would fix this too, since detail would be repressed due to proximity to the adjacent masked areas.

174 Messages

 • 

3.8K Points

10 years ago

Rob, LR Detail is the devil. I rarely go above 10. In fact, I rarely go above zero.

Push it up to max just to look at the grain structure it creates and tell me if it create shapes you want applied to to your pixels at any opacity.

Amount, radius, masking = pleasing.

Detail = trouble.

I like that image, btw.

Have you tried Detail and NR both off?

4.5K Messages

 • 

76.3K Points

10 years ago

Hey Scott,

Yeah, I hear you about the detail.

For this image, at such high ISO, if I were sharpening it soley in Lightroom, I would definitely turn detail off completely, and keep NR low - begrudgingly (NR does good things everywhere except the bee's fur, which if more than a tiny bit essentially ruins it). And then just hand sharpen the bee. In fact, here it is:

Global Sharpening
Amount: 0
Radius 1.3
Detail: 0
Masking: 0

Lum. NR: 8

Not as sharp as some of the others, and still has a fair amount of grain in the bokeh, but has a natural look without artifacts, IMO.

Final Thoughts:
--------------------
I usually sharpen *some* detail if its a low ISO shot (such that the detail is real, and not just sharpened noise), and/or the picture really warrants it: textured surfaces like rock... but, if I decide to cut corners and use the masking, I almost always crank the detail all the way down to zero, to lessen the sparkling.