BragIt
Image Fidelity

IT8.7 Profiling Tests — Update

This is an update to the article "IT8.7 Profiling Tests" — click here to go to the start of that article, which describes details of the setup and how the profiles were made etcetera.

Also in this update, it shows what multisampling can do to reduce noise levels.

[Stockholm, January 2006, updated February 2006]

Gamma 2.0 is better for the problematic image

Since the referred article was written in December 2004, I now recommend to do everything in gamma 2.0, also when inCamera profiles are used. The reason is that it provides a better base for lifting deep shadows. Whenever you convert from the input profile to the working space, such as Adobe RGB (1998), any color errors will be amplified, and they will be more so if the input profile is in gamma 1.0.

In addition, lifting shadows before converting to the working space seems to be a very effective way of avoiding profile artifacts in deep shadows, and this lifting seems to work better when the gamma is at 2.0 rather than at 1.0. These severe artifacts are probably due to the fact that the profile itself is very bad in the very darkest areas of the color space, simply because the IT8 target itself has no colored patches at all in the very deepest shadow regions — there simply is not much color data available to build upon! Lifting the shadows just a notch will take you away from the region of bad artifacts, so the color profile conversion can work with better quality data, resulting in much better shadow quality.

The inCamera profiles are now done on a target that has been multisampled. That is done to provide the best possible shadow quality of the profile. But beware: although this particular scanner, an ArtixScan 4000tf, has hardware support for multisampling, it still behaves very erratic — sometimes altering the brightness, sometimes it misses focus a bit, sometimes a different hue of near-black, but sometimes with excellent results!

In the hope for improvements, the firmware of the scanner was updated in February 2006 to version 1.6.1. The basic erratic behavior has unfortunately not disappeared, but possibly, the upgrade seems to have reduced the problems somewhat (it is difficult to know for sure). The erratic behavior does not only happen when using multisampling — different occasions of single-sample scannings can also vary noticably. For real perfection, one needs to do a few scannings and then pick the best one!

The original inCamera profile was made from a target that was not multisampled. Much later, (but before the firmware upgrade), I made a profile based on 16 times multisampling. After the firmware upgrade, I made yet another and better profile based on 8 times multisampling, by selecting the best and darkest scan out of a few scanning attempts. 16 times multisampling is usually much more than reasonable, since most of the improved noise suppression is achieved already with multisampling set at 2 times, but improvements can be seen up to about 8 times multisampling.

The examples shown below use the same problematic image as was shown in Comparisons of results for a problematic image. In addition, there are examples from new scannings of that image, using 8 times multisampling. Finally, comparisons are made of multisampling 2 and 8 times as compared to single-sampling.

Comparison methods

The images shown on this page are scanned using SilverFast with '48 Bit HDR Color', i.e as "raw" as possible. The gammas are 1.0 and 2.0 respectively, and no color profile is embedded from SilverFast. Rather than applying Curves to them to lift shadows, as was done in the beginning of the article series, the Photoshop adjustment 'Shadow/Highlight' is used instead. Just as in the other tests, prior to saving these images in a web friendly format, all of them have been converted to the sRGB profile. See start of article series for further details of that.

(As before, the profile is embedded in the images, so in Safari it will look right, but since most browsers on Windows don't understand color management, a correct judgement actually requires Windows users to save the images to disk and view them with a color management-savy program. However, all images here are in sRGB, so it should at least be consistent even if you look at them with an inferior browser.)


[1] Original image, gamma 1.0.
problematic image
Click to see a larger version in a separate browser window (92 kB).
This is exactly the same image (the very same scanning) as experimented with in Comparisons of results for a problematic image, i.e it is scanned in SilverFast at gamma 1.0 with the setting '48 Bit HDR Color', assigned the same profile that was made using inCamera 3.1 in December 2004, and then converted to sRGB.

Note the blue cast in the shadow!

[2] Original image, gamma 1.0. Lifted shadows > sRGB
problematic image
Click to see a larger version in a separate browser window (120 kB).
Same image as above, but we have applied the Photoshop adjustment 'Shadow/Highlight' at Amount 4%, done before sRGB conversion. This is reasonably close to the real slide. Note that the blue cast in the shadow is gone!

One bad thing with this procedure is that: after the shadows are lifted (while still in the input space) the following conversion to our working space (in this case sRGB) causes an enormous change in appearance! (Before conversion to sRGB, the shadows are completely blocked.) This means it is extremely hard to judge the effect of the shadow lifting, since we can't really see the result until after conversion to our working space.

[3] Multisampled image, gamma 1.0 (new inCamera profile)
problematic image
Click to see a larger version in a separate browser window (92 kB).
Same slide, but it is now multisampled 8 times and assigned the new profile that was made from a 16x multisampled IT8 target. This is also at gamma 1.0, and it should be compared with the "[1] original image, gamma 1.0" version above. Note that they are very similar!

(The profile made from 16x multisampled IT8 target, doesn't really make much of a difference compared with the original profile. However, the images are slightly different, and the reason is that it is a new scanning, and it is multisampled.)

[4] Multisampled image, gamma 1.0. Lifted shadows > sRGB
problematic image
Click to see a larger version in a separate browser window (108 kB).
Same image as above, but we have applied the Photoshop adjustment 'Shadow/Highlight' at Amount 4%, done before sRGB conversion.

Not very good, since there is a nasty blue cast in the darkest grains, and deep shadows are almost blocked. The reason it is different from the corresponding image [2] is not that the profile is different! It is the actual multisampled scanning that is different, and it seems to be so from time to time. This particular instance is dark (low noise), but color errors get amplified after conversion to gamma 2.0.

[5] Multisampled image, gamma 2.0 (new inCamera profile)
problematic image
Click to see a larger version in a separate browser window (96 kB).
Same slide, but multisampled 8 times at gamma 2.0 and assigned the new profile that was made from a 16x multisampled IT8 target (now at gamma 2.0).

It should be compared with images [1] and [3] shown above. Note that it has much less blue cast in the shadows, but on the other hand, the deepest shadows are blocked.

[6] Multisampled image, gamma 2.0. Lifted shadows > sRGB
problematic image
Click to see a larger version in a separate browser window (104 kB).
Same image as above, but we have applied the Photoshop adjustment 'Shadow/Highlight' at Amount 3%, done before the sRGB conversion.

Great! Very close to the real slide!

Besides this is the best result so far, one big difference with this procedure compared with doing the same thing in gamma 1.0 is that: after the shadows are lifted (while still in the input space) the following conversion to our working space (in this case sRGB) no longer causes such a big change in appearance! This means it is much easier to judge the effect of the shadow lifting.

[7] Multisampled image, gamma 2.0. Lifted shadows > sRGB > Curves
problematic image
Click to see a larger version in a separate browser window (200 kB).
Same image as above, but after converting to sRGB, we apply Curves 26->51 and clip black at 6 units.

This is not meant to be the most faithful representation of the original slide, but meant to show that this version of the image was good enough to be used for further editing in Curves without nasty colors or artifacts popping up. The shown tweaked image is about as much as can be sucked out from that slide — there simply are no more details in the real slide!

[8] Multisampled image, gamma 2.0. sRGB > massive lifted shadows
problematic image
Click to see a larger version in a separate browser window (208 kB).
This is just to show that one cannot achieve the same result as the previous image by applying shadow lifting after the sRGB-conversion. Here, massive amounts of shadow lifting is done: Amount 100%. Completely washed out and harsh!
[9] Multisampled image, gamma 2.0. sRGB > massive Curves
problematic image
Click to see a larger version in a separate browser window (260 kB).
Likewise, if shadows are not lifted before the sRGB-conversion, applying lots of Curves after the sRGB-conversion would also be terrible! In this example, Curves is 26->102. As usual with Curves, nasty colors start popping up.

All of the multisampled images above were using the profile that was made using a 16 times multisampled IT8 target (done before the scanner was firmware upgraded). Below is a test using the very newest and best profile, which is based on 8 times multisampling of the IT8 target, selected as the best and darkest scan out of a few scanning attempts. It actually makes a noticable difference!
[10] Multisampled image, gamma 2.0, assigned the newest profile
problematic image
Click to see a larger version in a separate browser window (100 kB).
This should be compared with image [5]. The only difference is that this one is assigned the very newest and best profile based on a selected 8 times multisampling of the IT8 target.

Notice how much better it is, with no blue cast and with much more detail in the shadows, since they are no longer blocked! The reason is that the IT8 target used as the basis for the profile was scanned as noiseless as possible with darker blacks, which enables the profile to better map colors from scans that are as dark as those in the target.

[11] Multisampled image, gamma 2.0, newest profile.
Lifted shadows > sRGB
problematic image
Click to see a larger version in a separate browser window (100 kB).
This should be compared with image [6]. Same image as above, but similar to what we did with image [6], we have applied the Photoshop adjustment 'Shadow/Highlight' before the sRGB conversion, but in this case we only need an Amount of 1% instead of 3% to achieve the same effect!

The reason that we only needed to apply 1% shadow lifting here is that the profile is different, and this version of the profile is better at representing the darkest shadows without pushing them down to almost pitch black.

In practice though, 2% shadow lifting would actually be best if we also adjust Curves somewhat to set the black point; see the next example [12]:

[12] Multisampled image, gamma 2.0, newest profile.
Optimally lifted shadows!
problematic image
Click to see a larger version in a separate browser window (148 kB).

Extremely close to the real slide! There is nothing more in that slide!

This is just like the previous image, but we have applied 2% shadow lift instead of 1% while in input space, but in return adjust Curves after the sRGB conversion so as to clip black to 6 units while locking the rest of the curve flat.

This version is the most accurate version of all, if the real slide should be the norm, although it is extremely similar to [6]. The jpg quality had to be increased from level 8 to level 9 in Photoshop so as to preserve the darkest area without objectionable color artifacts in the grains. This version is sufficiently good that it may be worthwhile to even study the grain structure in the darkest shadows, but to do that, the jpg compression must be reduced — click HERE to see the same image with jpg compression quality set at level 10 (232 kB).

[13] Multisampled image, gamma 2.0, newest profile.
Optimally lifted shadows > sRGB > Curves
problematic image
Click to see a larger version in a separate browser window (192 kB).

Extremely close to the real slide viewed on a very bright and good light table! There is nothing more in that slide!

This should be compared with image [7]. Same image as the previous one [12], but just as we did with image [7], we have applied Curves 26->51 and done some mild clipping (5 units to green and 6 units to blue, while locking the rest of the curve). This is the most accurate one relative to the real slide when it is viewed on a very bright and good light table.

This little 2% shadow lifting is very important, since without it, the shadows would get a nasty blue cast and be very blotchy.

Overall, it shows that the combination of suitable multisampling and an extremely carefully selected profile, and a slight lifting of shadows before conversion to the working space, extracts virtually everything from the slide, and provides a good basis for further manipulation. Even if the shadow lifting is only 2%, it is still very important!


Multisampling tests


The following sequence illustrate how multisampling affects the noise level. But as stated previously: beware that although this particular scanner has hardware support for multisampling, it behaves erratic as regards the darkest regions — sometimes altering the brightness, sometimes a different hue of near-black, but sometimes with excellent results. Each of the following examples is taken as the best scan of four scans, so as to roughly represent the best one can expect using: no multisampling, 2 times multisampling, and 8 times multisampling.

To clearly see the effect, these images have all had the Photoshop adjustment 'Shadow/Highlight' applied at Amount 2%, before the sRGB conversion, and then Curves 26->51.

[15] Single-sampled image
problematic image
Click to see a larger version in a separate browser window (248 kB).


Quite nasty RGB noise!
[16] 2 times multisampled image
problematic image
Click to see a larger version in a separate browser window (220 kB).


Much better!
[17] 8 times multisampled image
problematic image
Click to see a larger version in a separate browser window (180 kB).


Better yet! But the improvement can normally only be seen in rather extreme cases like this, when we concentrate on a large dark area that has some details way down in the shadows, and we choose to lift those shadows.

CONCLUSION
Lifting shadows before conversion to a working space can improve the result enormously! It avoids nasty artifacts at the very darkest portions of a scanner profile. Doing this adjustment works best at gamma about 2.0. But extreme care must be taken when making the profile. Multisampling clearly improves noise performance, up to about 8 times multisampling, although 2 times will usually be enough.

Back to the beginning of the article "IT8.7 Profiling Tests".

Harald E Brandt
Hägersten, Stockholm
Sweden
Photo: http://photo.bragit.com
BragIt: http://bragit.com
Email:
Last updated: 2011-10-27 at 15:07:10 +0200