RAW skin tone conversion comparison

[John_Nevill]

In an earlier thread, Asher mentioned that it may be intersting to see the effect that RAW converters have on skin tones, so i've obliged.

I used the default settings of each RAW converter, as shot WB and no added sugar. All files are natively cropped, resized and converted to HQ jpegs.

C1
http://www.eospix.com/jlnimage/RAWSkintones/C1.jpg
DPP
http://www.eospix.com/jlnimage/RAWSkintones/DPP.jpg
RSP
http://www.eospix.com/jlnimage/RAWSkintones/RSP.jpg
Silkypix
http://www.eospix.com/jlnimage/RAWSkintones/Silkypix.jpg

[John_Nevill]

Part two...

ACR
http://www.eospix.com/jlnimage/RAWSkintones/ACR.jpg
Bibble
http://www.eospix.com/jlnimage/RAWSkintones/Bibble.jpg

[Asher Kelman]

John,

Thanks so much for the valiant for the effort.

1. What is your own impression? This is important.

2. Can you optimize each one. But based on what standard.

3. How fragile are the files to further manipulation optimization and sharpening? What are the advantages of each/

4. How do the optimized images print.

4. How exactly are you posting the images here: uploading ot linking to a server? The latter is important with large files, which are needed here.

5. Could you make the original file available too?

6. This is a superb start. I presume you are using a PC or are you using Bibble from Virtual PC in Mac or on an intel Mac?

EVERYONE: Now, the constructive critique.

Let it begin!


Asher

[Andrew Rodney]

Time to add that sugar as at this point, it only tells us about default rendering which doesn't begin to tell a full, compelling story.

If you have access to a Mac, I'd love you to see how RAW Developer does as well.

[Asher Kelman]

Andrew, See my point # 3 for sugar!

But to what standard? How can one set guidelines. It is a lot of things, but not limted to gamut, fragility, roundings off, detail, noise etc.

Or does one simply make a geshtalt "my best feeling" optimization and do that for each one. As an artist, that is fine, but scientifically it is troubling.

All I know is that the color of Canon's own translations is said to use data, (according to Chuck Westfall I think, that may not be accessable by other RAW development programs. Also the word is that Silkypix gives the best skin tone.

"As is" processing is of no value to me, like a bride without makeup would never get married!

Still this is an excellent start. Can we have the RAW file to do more work with?

Asher

[Andrew Rodney]

I've been spending all kinds of time trying to compare the same RAW files with multiple RAW converters. NOT easy. In fact, almost impossible to get a baseline rendering to compare (do you use or not use sharpening or noise reduction?). It's tough. However, just looking a at a default rendering only tells you a fraction about how the converter may preform. The best converters allow you to at least tweak settings to your liking and build one or more defaults for all other files.

There is no standard. That's the blessing and curse of RAW. Sure, you can look at things like ugly artifacts (Aperture 1.0 took honors). In the end, the rendering YOU prefer is what makes one converter better than another (of course workflow comes into play although the quality of the conversions in my mind is key).

A converter that makes the best skin tone might really suck at other imagery.

I've been emailing two esteemed geeks on the ICC digital camera committee (Jack Holms of HP and Eric Walowit) about a means of testing RAW converters in a totally objective way. Do you shoot a Macbeth and measure the colorimetrically correct color of from converter. Both told me there's no way to accomplish this. Assuming you produce a colorimetrically accurate rendering of a Macbeth, that might produce a butt ugly rendering of a totally different image (scene). They both told me that (in a nutshell) the RAW converter YOU the image creator prefer is the "best" converter.

The ability of a RAW converter to produce a rendering you like usually takes more than simply pointing the converter to a RAW and asking for a default setting. It's a start but just a start. What's the potential of the converter to totally alter that rendering to produce a skin tone you like?

[Asher Kelman]

Well then Andrew, what one needs to do now is then download these files, optimize to taste and print.

It would be bettter to do this at the outset!

Asher

[John_Nevill]

One could no doubt tweak them all to look similar e.g. level of contrast, saturation and WB etc. But surely that would be subjective? What would I choose as the baseline?

All these are "out of the tin", simply loaded, cropped, resized and converted. I suppose one could say thats how the respective development teams percieved the renderings to be.

I use an Acer Duo Core t2500 laptop with ATI X1600 dual head gfx card, eye one calibrated dvi Eizo Flexiscan LCD.

So what's my perception, apart from significant variation/interprtation of the same dataset?

ACR is natural and slightly warm

Bibble pushes upper mid tones, brighter and is less contrasty.

C1 yeilds more contrast and has a slight orange hue

DPP is a strange one, I used standard colour settings which tends to desaturate skintones.

RSP has similar colour to C1, yeilds greater contrast (punch) and is less smooth and a little unkind to skin.

Silkypix seems to be colour similar to ACR but with more contrast and almost waxy looking. Very forgiving!

In this portrait example, IMO, Silkypix and ACR have the edge. However, I'll spend some time optimising using ACR as the baseline. It would be nice to see if one can achieve equality.

I'd also like to do the same with a wide dynamic range landscape and see how clipping, saturation and hue are handled. Something tells me that i'd prefer RSP.

Unfortunatley the RAW file is not mine its a colleagues, however i'll ask him.

BTW, the files are all links to my host server. So no OPF overhead.

[John_Nevill]

Andrew, I spent a rainy weekend during winter 2005 trying to level sharpness parameters across RAW converters.
I used Imatest to evaluate TIFF outputs and went through an iterative process of level adjustment until they matched.
Here's my findings! (http://www.eospix.com/content-30.html)

[scott kirkpatrick]

In your C1 and RSP examples, did you use the program generic or ETC profiles, or ...?

I've done your sharpening study comparing different betas and releases of C1 and early versions of RSE, with results consistent with yours. Yes, the hidden differences are dramatic. Wouldn't it be great if RAW converters came with actual absolute parameters, rather than arbitrary numbers?

scott

scott

[Asher Kelman]

John,

that is in interesting set of data. So theoretically, one could create for say faces and landscape and buildings, for example, sets of equalizing parameters and then use these to produce images of the same sharpness, color contrast etc and now see how fragile the images are and in what way images show other artifacts.

BTW, I had a problem with your use of the word "waxy" with respect to skin and feeling it was good, unless you are the mortician!

Asher

[John_Nevill]

John,

BTW, I had a problem with your use of the word "waxy" with respect to skin and feeling it was good, unless you are the mortician!

Asher

I dared not call it silky :o)

[Asher Kelman]

So let's go with silky.

Asher

[John_Nevill]

In your C1 and RSP examples, did you use the program generic or ETC profiles, or ...?


All done with internal ICC profiles.

[David Miller]

Well, I guess I'm just a numbers kind of guy, so I took all six files into PS for a look. Noted that they were all sRGB except the DPP file which was Adobe1998. Converted it to sRGB before making comparisons. The histograms are interesting; in RGB the green and blue channels are quite similar in shape for all six files - the big differences are in luminosity level; the red channel shows significant differences in shape between files, in addition to luminosity differences. The differences in luminosity confound visual attempts at hue discrimination. I took the files into LAB for some measurements. The model's eyes, and hair, are nearly identical in color across all six files - perceived differences are due primarily to luminosity (contrast and level) differences. The greatest difference is in the rendering of the lips. Measuring the same location on the lips in all files, all but C1 have the same level of yellow (C1 is higher than the rest). Silkypix is the most magenta, DPP the least. Measuring the same location on the cheek in all files, all have roughly the same level of yellow. RSP is the most magenta, DPP the least. Rank ordering the converters for magenta level from greatest to least: for the lips - Silkypix, Bibble, RSP, C1, ACR, DPP; for the cheek - RSP, Silkypix, ACR, C1, Bibble, DPP. The luminosity levels of the files could be brought into alignment with a tonal curve adjustment. Color is more problematic, they can be brought closer (an infusion of magenta to DPP for instance), but I doubt an exact match could be attained. I would be interested in knowing which lip shade is most accurate. I'll bet the model could tell us (with a degree of accuracy exceeding the most sensitive colorimeter), or I don't know women!

Dave

[John_Nevill]

David, the DPP conversion in Adobe colourspace intrigued me. So I revisted it and I indeed I set the wrong working colourspace. I have since changed the file.

[Asher Kelman]

We're getting closer to understanding this!

Great work John and David.

So we might be able to describe the, at least initially by their magenta richness and luminence histiograms?

Could we ask the model?

I think, to do this on with a color card, one would need thousands of colors.

Also for the gray scale card we could look at that too.

Is the model available to ask her to show you the lipstick?

Asher

[Michael Tapes]

What is important here to me, is a point that I have been trying to make to many people as they compare the "color" output of different converters. The perceived color is dictated by many factors other than color. In other words, as Dave pointed out, one can measure the same color in Lab space, yet the perceived color is different based on tonal differences, contrast of the image, etc. Saturation will of course also change the color. So when many people compare and say that this output has better color than that output file, then often times are being influenced by other factors, and the correct "fix" for "bad" color is often not a new camera profile or adjusting the WB, but making a tonal adjustment, using curves, or contrast or saturation controls.

It is like my old audiophile days. Someone would say how much they "hated" those speakers when listening to a supposed high quality system. Of course they would have no way to know that it was not the amplifier, or turntable, or preamp causing the bad sound. In fact the better the speaker, the worse the sound will be if something in the previous chain is sub-standard, including the recording itself, or the storage format (early Cds, for example). The analogy in the digital photo world, is the color is wrong when in fact it is another parameter making the color appear wrong.

So what's a mother to do.
Change the speakers, or find the issue and correct it.
Change the raw converter, or camera profile, or White Balance, or know how to use the tools to get to what you want. of course the tools that get you there the fastest are the ones to use for you. For someone else it might be a different tool set.

[Asher Kelman]

I think that the convertors might have different value for different subjects. I would look to use another program if I wan't getting what I needed easily with the first program.

So far, I like the look of SilkyPix.

Asher

[jkardell]

I think that the convertors might have different value for different subjects. I would look to use another program if I wan't getting what I needed easily with the first program.
Asher

Are these raw converters actually using slightly different algorithms to calculate color and luminosity, or, is it simply a matter of different default settings?

I am very ignorant of the inner workings of raw converters, so please bear with me a moment. Is the Bayer demosaicing algorithm a "standard" or actually a more malleable process that can be "tweaked" for particular sensor characteristics. If the latter, then the variability of the algorithms may result in a situation where you might not be able to generate identical output from two different converters.

Fascinating thread, thanks!

Best regards,
Jonathan

[Gary Jean]

Additional complications....how to compare conversions for different camera models?

[Peter Mendelson]

When I was scrolling down the samples at the beginning of the thread, I immediately knew which one was from RSP without seeing the labels of the converters. I really would like to use RSP and bought RSP Premium plus Color Engine and Mike Chaney's profiles, but when I tried Silkypix I found I was getting much nicer output from my 5D without much effort. I have and use a Whibal card, and really like the workflow of RSP, but for me there really is no comparison - I have to work too hard to get RSP output anywhere near as nice as what I get from Silkypix. Maybe others can do it, but I can't.

I know there was talk on the Pixmantec forums about having different modes so that RSP can default to different looks right out of the box. That may be anathema to those who like the muted files that RSP starts with so they can tweak to their hearts' content, but personally speaking, I am staying away from RSP until they can do something like this. For example, when you first import raw files into RSP, I believe you see the JPEG thumbnail version of the file for a few seconds until it turns into the unprocessed Raw file. If RSP can give you the option of providing that JPEG look or variations thereof (like DPP Portrait, Faithful, etc.) as alternatives to what it does to the Raw file now, I think RSP will only attract and keep more users. For me, the Color Engine low and high sat profiles still don't get there.

Don't get me wrong, I like to tweak curves, color settings, etc. too and want to gain maximum value from shooting Raw, but Silkypix seems to provide me with a better starting point and allows me to get to the end point much more quickly.

Just my 2 cents.

Peter

[John_Nevill]

Ok to add some weight to MT's comments, which is which?

One's a custom ICC profile in RSP the other's Silkypix.

http://www.eospix.com/jlnimage/RAWSkintones/A.jpg

http://www.eospix.com/jlnimage/RAWSkintones/B.jpg

[Jon Mark]

Magne's profiles for C1 is known for improvements in skin tone compared to the generic profile. I'd like to see the same image with Magne's profile in C1 compared with Apple's Aperture...

[John_Nevill]

Here's the RSP CE (Magne) version. Unfortunately I don't have a Mac nor Magne's C1 Profile.

http://www.eospix.com/jlnimage/RAWSkintones/RSP-CE.jpg

[Michael Tapes]

Are these raw converters actually using slightly different algorithms to calculate color and luminosity, or, is it simply a matter of different default settings?

I am very ignorant of the inner workings of raw converters, so please bear with me a moment. Is the Bayer demosaicing algorithm a "standard" or actually a more malleable process that can be "tweaked" for particular sensor characteristics. If the latter, then the variability of the algorithms may result in a situation where you might not be able to generate identical output from two different converters.

Fascinating thread, thanks!

Best regards,
Jonathan

Yes, each converter uses proprietary algorithms for each phase of the process. The exception would be RCs that use an SDK like BreezeBrowser does for the canons, or people that have "lifted" generic code from David Coffin (dcraw). But speaking of Pixmantec and Capture One for sure, they are proprietary algorithms developed, in this case both by Michael Jonnson, but each of those programs is unique as well. I cannot speak for the other converters.

[Michael Tapes]

..snip..
but personally speaking, I am staying away from RSP until they can do something like this.
Peter

I predict that you will see a nice change in RSP/CE in the near future. Then again, I could be wrong...

For those who are unaware, CE stands for Color Engine, which is an optional "plug-in" to Raw Shooter, which is a set of Camera Color profiles for certain CE supported cameras, that can be tweaked by the "engine" part of the CE to produce varied camera color mappings.

[Joe Marques]

Here's the RSP CE (Magne) version. Unfortunately I don't have a Mac nor Magne's C1 Profile.


John, if you send me the RAW (email me or www.yousendit.com) I'll run it through C1 with Magne's profile. I've done my own 7-way RAW skin tone comparison - white balance set with WhiBal (I tweaked each image to try to achieve what I perceived as a pleasing result - previously I had just run files through with default settings):

http://www.joemarques.com/upload_images/5D/5WAYRAW/7-wayAdjust/

I find C1 with Magne to be superior for skin tones - best results with least amount of effort/trouble.

[Michael Mouravi]

Another thing is that each converter behaves differently with different cameras. For instance, with my original 1D, C1 and DPP produce best results, while RSP's defaults are pretty bad and even after tweaking, I prefer the C1.

With 1D Mark II N, RSP's defaults seem very good in my limited experience, not really far from C1 and DPP. I haven't had a chance to see how the converters perform at settings other than default with MarkII N's files...

[Peter Mendelson]

Michael, thanks for your reply, and I do hope to see changes in RSP soon. Just to do an informal test for myself, I converted a Raw file from my 5D using Silkypix and RSP with CE and Magne's low sat profile. I used the "auto" settings for Silkypix, and prefered the output greatly to what I could achieve in RSP, whether I used the auto settings or tweaked RSP as well as I could (I didn't need to tweak Silkypix to get a look I liked). I tried using color temp, saturation, vibrance, curves, various contrast settings, etc. in RSP, but I just couldn't get as nice a result. You can blame me for not having adequate skills at using RSP, and I am not an expert, but I did use it for several months and it just shouldn't be this hard for someone like me to get nice results. You or others may even prefer the RSP results, but I find that when I try to warm up the skin a bit I get too much orange and red and I can't get a really nice balanced tone.

If anyone has any specific recommendations for getting really nice skin tone, I would love to hear them. If anyone wants to try adjusting this raw file themselves with RSP to show how to get better results, please let me know how I can send it to you.

Thanks! Peter

Here are the results (links to large photo):

Silkypix auto:

http://www.zenfolio.com/img/p477328582.jpg

RSP auto:

http://www.zenfolio.com/img/p252970015.jpg

RSP tweaked:

http://www.zenfolio.com/img/p308581236.jpg

[Michael Tapes]

Peter,

I would only suggest from a forum perspective to use smaller versions of the pics so they download faster and can be viewed in full. When viewing color issues, it is easier and faster to view smaller files IMHO.

Thanks for the post and examples...

[Peter Mendelson]

Sure - I hope this better (although for people who really want to see the details they may want to view the larger shots I posted in my earlier email):

Silkypix:

http://peterm1.smugmug.com/photos/73686434-L.jpg

RSP with tweaks:

http://peterm1.smugmug.com/photos/73686524-L.jpg

Peter

[Don Lashier]

The thing is that any number of conversion settings can have a much more dramatic effect on skin tone that the difference between RC's. I've found that if you equalize tonality and WB, output from the various RC's is remarkably similar. Skintone seems to be particularly sensitive to both tonality and WB. See this example, all done with C1, but moderate variations in WB controls and tonality controls:
http://www.lashier.com/home.cfm?dir_cat=27442

- DL

[Bev Sampson]

Peter, I am new to RSP and like it very much. What tweaks did you make to the RSP file?

BS

[Michael Tapes]

The thing is that any number of conversion settings can have a much more dramatic effect on skin tone that the difference between RC's. I've found that if you equalize tonality and WB, output from the various RC's is remarkably similar. Skintone seems to be particularly sensitive to both tonality and WB. See this example, all done with C1, but moderate variations in WB controls and tonality controls:
http://www.lashier.com/home.cfm?dir_cat=27442

- DL

Don,

Perfect example of how Exposure, Contrast and Saturation affect the perceived color. people discuss changing the "color" via profile or WB, when in fact they could adjust the other parameters and have more control over the perceived color of the photo. of course WB and Tint "do" affect the true color, and is relevant as well. But easier to use a proper WB and adjust other things to keep the pixels in order.

Great to have you here!

[John_Nevill]

I totally agree with your findings, and yes one could do the same in RSP, silkypix etc, but I must add that there are noticeable differences in how each render engine interprets RAW data. Some exhibit patterning and mottling in their rendering algorithms, while others dont.

[Don Lashier]

I totally agree with your findings, and yes one could do the same in RSP, silkypix etc, but I must add that there are noticeable differences in how each render engine interprets RAW data. Some exhibit patterning and mottling in their rendering algorithms, while others dont.

Hi John,

I totally agree. I didn't mean to imply that choice of RC's doesn't matter, just that a lot of attributed differences of color, skin-tone etc. are in fact due to differences in default tonality treatment, which is where RC's most obviously differ re image appearance.

Similarly, claimed signifancant differences in shadow detail or highlight detail and recovery are also mostly due to different default tonality.

However, as you point out, there are often significant differences in how RC's handle shadow noise, aliasing (tight patterning), clipped highlight channels etc. I have some (rather outdated but relevant on principle) examples of this on my website.

In addition, the RC's choice (or choices) of default tonality treatment can also make it easier or harder to achieve the results you desire, and the answer is not necessarily the same for everyone varying with your total workflow and nature of images you shoot, and even brand/model of camera.

The toolset, workflow, and adjustment facility of RC's are equally important criteria.

- DL

[John Burkus]

I downloaded these RAW comparisons and have been comparing them on a screen that shows them all at one time. My conclusion is that I could live with any of these renderings because they are all very good. I too have spent countless hours comparing seven different RAW converters with different scenes and have concluded that by a large there is no one convereter is best for all shots.

I think that my problem is that I am a RAW converter junkie and I fear that there is no cure for it. I can hardly wait for my next fix. DxO Optics is coming out with a new converter is Sept. Nikon just came out with NX which my machine won't load. Lightroom for Windows should be around soon...and on it goes. Oh yes, that reminds me that I should be concentrating on taking better pictures rather than seeking the RAW Holy Grail.