16 Replies Latest reply on Sep 9, 2009 8:36 AM by Bill_Janes

# The sharpening defaults, just for curiosity.

It's obvious that a 5 MP camera needs different sharpening setup as a 24 MP camera.

How can the ACR sharpening defaults to be always 25-1-25 for all cameras?

Thank you

Marco

• ###### 1. Re: The sharpening defaults, just for curiosity.

The engineers spent a fair while optimising the sharpening parameters (actually all settings) for a various range of camera model / sensor size. These parameters were then normalised so that we have a standard value for every supported camera model. That being said, you can fine tune the settings to suite yourself then save them as the camera default.

• ###### 2. Re: The sharpening defaults, just for curiosity.

Ian Lyons wrote:

The engineers spent a fair while optimising the sharpening parameters (actually all settings) for a various range of camera model / sensor size. These parameters were then normalised so that we have a standard value for every supported camera model.

This mean that 25 1 25 are not absolute values, but a percentage of those parameters?

Thank you

Marco

• ###### 3. Re: The sharpening defaults, just for curiosity.

Marco N. wrote:

This mean that 25 1 25 are not absolute values, but a percentage of those parameters?

It would not be accurate to use the term percentage, no. The "default" of 25 for any camera is adjusted on a model by model basis. Also, the parameter that is normalized (adjusted for differences) is really only the Amount setting. Radius is radius and Detail is Detail (which relies heavily on radius).

• ###### 4. Re: The sharpening defaults, just for curiosity.

Jeff Schewe wrote:

What I can not understand is how a radius of 1 pixel can be OK for any camera? For example, in PS sharpening Bruce suggested 1 pixel for a 4 MP camera and 0.4 pixel for 11 MP.

Thank you

Marco

• ###### 5. Re: The sharpening defaults, just for curiosity.

Marco N. wrote:

What I can not understand is how a radius of 1 pixel can be OK for any camera?

Because there is not a 1:1 relationship between Camera Raw and Photoshop. First off, Camera Raw does a luminance blend in linear space. Photoshop (unless you make a special gamma 1 color space) will ALWAYS have a gamma adjustment. Also, the way Camera Raw is applying the sharpening with the amount and detail is totally different than Photoshop (whch doesn't have any of that).

You would do well to essentially "forget" anything you think you know about Photoshop's Unsharp Mask filter as it applies to Camera Raw-cause they are totally different tools.

• ###### 6. Re: The sharpening defaults, just for curiosity.

Marco N. wrote:

What I can not understand is how a radius of 1 pixel can be OK for any camera? For example, in PS sharpening Bruce suggested 1 pixel for a 4 MP camera and 0.4 pixel for 11 MP.

I think you are referring to Table 5-1 (Page 167) of Bruce's sharpening book. That table is for sharpening for source, which takes the megapixel count into account, but does not consider image content-- low frequency or high frequency. In general, one uses a smaller radius for high frequency images with a lot of fine detail and a higher radius for low frequency images. Sharpening for source and content are often combined in capture sharpening and one must take into account both the megapixel count and image content. Of course, a high megapixel camera can capture more fine image detail.

• ###### 7. Re: The sharpening defaults, just for curiosity.

Jeff Schewe wrote:

Because there is not a 1:1 relationship between Camera Raw and Photoshop. First off, Camera Raw does a luminance blend in linear space. Photoshop (unless you make a special gamma 1 color space) will ALWAYS have a gamma adjustment. Also, the way Camera Raw is applying the sharpening with the amount and detail is totally different than Photoshop (whch doesn't have any of that).

You would do well to essentially "forget" anything you think you know about Photoshop's Unsharp Mask filter as it applies to Camera Raw-cause they are totally different tools.

For the standard unsharp mask the radius can be precisely defined mathematically, as described in this link. The radius is the standard deviation of the radius used for the Gaussian blur used to build the unsharp mask used for the sharpening. The blurred image is then subtracted from the original image and the result is normalized. Such a subtraction operation enhances high-frequency spatial detail at the expense (attenuation) of low-frequency spatial information in the image. If the blur radius has a standard deviation of 1 pixel, then 95% of the Gaussian blur radii would be between ± 2 pixels. In other words, a radius of 1 pixel reaches further than 1 pixel to either side of the edge being sharpened.

Does the radius of the ACR sharpening function have a similar definition? By the way, the Java demo on the link has a nice preview function demonstrating the results of the sharpening parameters for various types of images.

• ###### 8. Re: The sharpening defaults, just for curiosity.

Bill_Janes wrote:

Does the radius of the ACR sharpening function have a similar definition?

That's am "Eric or Thomas" question and I don't think there's an "easy" answer...it's certainly not as simple as USM.

• ###### 9. Re: The sharpening defaults, just for curiosity.

Hi Bill, there is not a direct, analogous definition for the CR/LR sharpening, since the CR/LR sharpening consists of multiple methods (or you can think of it as one big method comprising many steps).

Eric

• ###### 10. Re: The sharpening defaults, just for curiosity.

Hi Bill, there is not a direct, analogous definition for the CR/LR sharpening, since the CR/LR sharpening consists of multiple methods (or you can think of it as one big method comprising many steps).

Eric

Thanks, Eric, but I was hoping for a few more details. I understand that the ACR sharpening is based on Bruce Frasers sharpening workflow, which is described in great detail in his sharpening book, which soon will coming out in a revision by Jeff Schewe. Bruce's basic tool is the unsharp mask operating through a layer with blend if options to restrict the sharpening to the midtones and with an edge mask to restrict sharpening to the edges.The sharpening can be further modified by adjusting the opacity of the sharpening layer.

In Bruce's workflow, he ran the unsharp mask at a gamma of 2.2 or 1.8 and adjusted the blend if sliders appropriately for the gamma of the working space. I had assumed that ACR used an unsharp mask with a radius and amount similar to the unsharp mask in Photoshop. Jeff tells us the sharpening with ACR is applied to linear data. With ACR the edge mask is optional and the detail slider is based on totally new technology to filter out the radius halos on high frequency data while allowing the radius value to be applied to low frequency data as lower values for detail are used. At the maximum, detail allows the sharpening radius to be applied to all frequencies. Is this roughly true?

• ###### 11. Re: The sharpening defaults, just for curiosity.

Bill_Janes wrote:

I understand that the ACR sharpening is based on Bruce Frasers sharpening workflow, which is described in great detail in his sharpening book, which soon will coming out in a revision by Jeff Schewe.

While it's true that it's based on Bruce's sharpening workflow principles, it would be wrong to presume that ACR's sharpening is anything other and a pretty much new set of algorithms...Bruce's book was used by Mark Hamburg in determining certain aspects of what should be incorporated into Camera Raw's capture sharpening, but Mark has this habit of taking feedback and going off on completely new directions–which is exemplified in the "Detail Slider". You're description is ok...but how it was created is a bit more complicated. See, at one point I think there were 8 or 9 separate sharpening parameters–which Thomas thought was too many. He agreed to allow a total of four. So, Mark took about 3 sliders and combined them into a single slider adjusting a bunch of parameters all at once. On one hand, Detail is a halo suppression and on the other hand Detail emphasizes/de-emphasizes certain frequencies...all based (I think) on the main set radius–which while it is (I think) based on a blur, since the blur currently is in linear, the nature of the radius isn't 1:1 to Photoshop. I think Mark said something about it being a 1.3 or 1.4 factor (with LR's supposed to be smaller).

Camera Raw's sharpening also rolls off the highlights and shadows using it's own blend logic. But the EXACT numbers and manner of blend are considered proprietary...which is really the problem here. It's all well and good to ask questions directed toward how to use ACR's sharpening...but at a certain point stuff may cross a line between need to know for use and secret sauce. And, there's a lot of secret sauce in ACR's sharpening...

• ###### 12. Re: The sharpening defaults, just for curiosity.

Jeff Schewe wrote:

It's all well and good to ask questions directed toward how to use ACR's sharpening...but at a certain point stuff may cross a line between need to know for use and secret sauce. And, there's a lot of secret sauce in ACR's sharpening...

Jeff, thanks for your insight. Thomas, Eric, and Mark have done an excellent job implementing the sharpening in ACR, and I understand that you also played a significant role in the process. One does not need to know the internal workings of the sharpening workflow to use it effectively, and I would not expect that Adobe would divulge trade secrets. One situation in which one probably should not use ACR sharpening is with high ISO shots where the ACR noise reduction is not adequate and an industrial strength algorithm such as Noiseware is needed. One would not want to sharpen the noise before reducing it.

With Noiseware, one can use its built in sharpening or sharpen later. Ideally one would use layers with a surface mask for NR and an edge mask for sharpening. It would be nice if ACR had better NR so that everything could be done in ACR without needing Noiseware and going to all this trouble.  I am told by experts that NR should ideally be done early in the workflow, preferably during the raw conversion. Personally, I have not had much luck making edge or surface masks with noisy images. Perhaps the best that can be done is to mask off areas that do not need sharpening or NR respectively in the workflow. However, NR may not be a high priority with Adobe, since noise is usually not a problem with the latest dSLRs. Some guidance that does not reveal trade secrets would be appreciated by many of us for use in these difficult situations where we are on our own.

• ###### 13. Re: The sharpening defaults, just for curiosity.

Bill_Janes wrote:

One situation in which one probably should not use ACR sharpening is with high ISO shots where the ACR noise reduction is not adequate and an industrial strength algorithm such as Noiseware is needed. One would not want to sharpen the noise before reducing it.

Yeah ya know I actually kinda bought into that as well. In point of fact since ACR 4.4-4.6 I think the noise reduction in Camera Raw when PROPERLY combined with sharpening has gotten a bad rap. In the most recent book I compared ISO 1600 Noiseware Pro and Camera Raw 5.4 and found them really pretty close. The key is of course to properly adjust BOTH the sharpening (including the Detail slider) and the noise reduction. Yes, I still use Noiseware Pro when all other methods fail....and yes, Noiseware Pro is better than anything else inside of Photoshop (just beating out Surface Blur with an edge mask). But, if you are starting with "reasonable" digital camera noise from a "real" ISO (not the over-amped numbers) then Camera Raw can do quite well. I've even had good success doing initial sharpening and noise reduction inside of Camera Raw and then doing additional noise reduction targeted primarily at deeper tones (using blend if) with Noiseware Pro on top of the ACR sharpening and noise reduction.

All that said, I'm pretty sure that there's more image quality left in digital captures with even further improvements in Camera Raw....the engineers never rest ya know? All this stuff keeps getting better and better!

• ###### 14. Re: The sharpening defaults, just for curiosity.

Jeff Schewe wrote:

Yeah ya know I actually kinda bought into that as well. In point of fact since ACR 4.4-4.6 I think the noise reduction in Camera Raw when PROPERLY combined with sharpening has gotten a bad rap. In the most recent book I compared ISO 1600 Noiseware Pro and Camera Raw 5.4 and found them really pretty close. The key is of course to properly adjust BOTH the sharpening (including the Detail slider) and the noise reduction.

PROPERLY is sometimes difficult to achieve in the real world, so I just ordered your book in order to gain some understanding of what is available with ACR. Since I have been using the Nikon D3, noise is less a problem and I hope that proper use of ACR will be sufficient for most images and Noiseware will not be needed. Many times a bit of noise in the image is preferable to the softening introduced by noise reduction.

• ###### 15. Re: The sharpening defaults, just for curiosity.

Hi Bill, unfortunately I cannot provide you any additional details about the CR/LR sharpening methods. I understand that you are curious, but as both you and Jeff noted, there is only so much we can share. Of course, it is important that you can extract the quality you're looking for, so if you have specific questions about the workflow and/or suggestions for improvements, feel free to fire away. You should find Jeff's new book helpful ... Particularly since whole sections have been added / revised / rewritten with the CR/LR/PS workflow in mind.

• ###### 16. Re: The sharpening defaults, just for curiosity.

Hi Bill, unfortunately I cannot provide you any additional details about the CR/LR sharpening methods. I understand that you are curious, but as both you and Jeff noted, there is only so much we can share. Of course, it is important that you can extract the quality you're looking for, so if you have specific questions about the workflow and/or suggestions for improvements, feel free to fire away. You should find Jeff's new book helpful ... Particularly since whole sections have been added / revised / rewritten with the CR/LR/PS workflow in mind.

While waiting for Jeff's book to arrive (it has been shipped by Amazon), I performed a few tests to gain some insight into ACR sharpening parameters and am posting the results for discussion.  I photographed an Imatest chart with the Nikon D3 and performed some analysis with Imatest. Imatest documentation is on their web site, but briefly for those not familiar with the program, it uses a slanted edge target and looks at the transition from dark to light. The slope of the rise from 10% to 90% brightness should be as high as possible without producing artifacts.

Here is the Imatest plot of an ACR rendering with no sharpening with the inset image showing a 100% view of the actual image being analyzed as rendered by ACR:

And here is the plot of the default ACR sharpening using a Radius of 1, amount of 25 and detail of 25. The slope is improved and there are no objectionable artfacts.

http://bjanes.smugmug.com/photos/644043506_bcCf8-O.png

For comparision, this is the same image with no sharpening in ACR and with an unsharp mask applied in Photoshop using a radius of 1, amount 100% and threshold of 0. The slope is improved, but the overshoot shown on the plot results in a halo at the black-white interface shown by the Imatest preview of the interface (note the magenta cross hair) and in the test chart image.

http://bjanes.smugmug.com/photos/644043523_8Y2vi-O.png

For comparison, here is a plot of the ACR sharpening with a radius of 1, amount 100, and detail 100. The marked overshoot produces a rather pronounced halo and there are severe artifacts.

http://bjanes.smugmug.com/photos/644043476_PdCHw-O.png

Reducing the detail to the default of 25 but leaving amount at 100 eliminates the overshoot and reduces the halo. IMHO, this is the best compromise for this image among those shown, but further adjustments would likely be helpful.

http://bjanes.smugmug.com/photos/644043493_f8vSC-O.png

This exercise helped me to understand the adjustments and hopefully will be useful to others. The MTF values with excessive sharpening should be ignored. Imatest can not distinguish alaising artifacts from true detail and the reported MTF is above the Nyquist limit of the sensor.