Veiled Bride: Results


Dan Margulis
 

I’ve posted the first pass of our Veiled Bride study as described below.

 

Reviewing: This is one of 5,000 images taken from the MIT study, of which 100 were randomly chosen for my own testing. Five retouchers were paid to have at each image, getting a .dng file to work with. The group did notably badly on this one; only two got marginally acceptable results, the other three were disasters.

 

I therefore gave the image out as an exercise in my ACT class in 2018, expecting better results. It didn’t happen. Personally I don’t find the image all that hard, but a lot of people evidently do, which is why I picked it for this group.

 

We have 22 entrants. When a person submitted two versions, I chose the one I thought was better. Everyone also submitted a list of their steps, thanks very much. I haven’t read these, because I’d rather get a sense of who was successful and who wasn’t before investigating why. 

 

The files don’t have people’s names on them, and were random-generator numbered from 301 to 322. As with our Carnival study, we also have a “par” version, #323. To get it, I chose what I thought looked like the five best entrants, and I averaged them, each one weighted 20%. This often creates a version that is superior to most if not all of its parents.

 

I’m going to have some things to say about this assortment, you betcha, but before the bloodshed begins it’s better IMHO to hear from the rest of you. Why do you think this one is so hard? What do the successful versions have in common? Meanwhile, if you’d like to know how your own version stacked up, download the par version and compare the two directly. Do you think you got the same kind of quality? If not, I hope you’ll find further discussion useful.

 

The folder is in the group Photos section, named Case Study: Veiled Bride. 

 

https://groups.io/g/colortheory/album?id=247497

 

Because I have a feeling that some of us would like a closer look at these, I also have zipped all 23 and uploaded a 35 mb file to our Files section,

https://groups.io/g/colortheory/files/

 

Search for Veiled_bride_entries_060120.zip

 

I look forward to your comments.

 

Dan


Gerald Bakker
 

Dan, first of all: the link you posted is incorrect, it brings me to the album of the previous exercise. Here is the correct one: https://groups.io/g/colortheory/album?id=247497
(Yours looks the same, but in reality it brings us to a different album id).

Interestingly, somehow this image looked familiar to me, so I searched through the old MIT blog entries and through my own archive but couldn't find it anywhere. Now I understand: I was in that 2018 ACT class. But I completely forgot how I did then, and how my correction compared to the other students' versions.

So why is it so difficult? Honestly, I didn't find it that difficult at all while working on it, but now that I've seen the other entries, it becomes more clear. There are a few pitfalls.

  • First, this image needs a good foreground/background distinction. Not every version succeeds in that respect. In #310, both are light which to me doesn't work well. In some versions like #307 and #320 the background is too colorful which I find distracting.
  • Second, it seems that the image cannot stand a color cast very well, presumably the (white) veil is especially sensitive for that. In #321 I see a bluish cast, #301 is too red.
  • Third, skin is always sensitive to small aberrations. In #319 it has too much detail, in #316 the face is a bit lifeless, and in #311 overly colorful.
There is much more variety in the submissions than in the Carnival exercise. It's easier to pick out favorites - on the other hand, there are a few that I really don't like. The par version is REALLY good, arguably the best of all. Well... it's not that far away from my own entry...

Question: can we eventually see the five MIT corrections?
--
Gerald Bakker
http://geraldbakker.nl


Dan Margulis
 



On Jun 1, 2020, at 11:45 AM, Gerald Bakker <gc.bakker@...> wrote:

Dan, first of all: the link you posted is incorrect, it brings me to the album of the previous exercise. Here is the correct one: https://groups.io/g/colortheory/album?id=247497
(Yours looks the same, but in reality it brings us to a different album id). 

Thanks for the correction.

Interestingly, somehow this image looked familiar to me, so I searched through the old MIT blog entries and through my own archive but couldn't find it anywhere. Now I understand: I was in that 2018 ACT class. But I completely forgot how I did then, and how my correction compared to the other students' versions. 

It’s probably best forgotten. #305 was the best student version from that class, and it’s no prize IMHO.

Question: can we eventually see the five MIT corrections?

The MIT group by and large bombed on this one, so much so that we can’t even look at their average. Three of them did so badly that it would be a waste of space to show them. The two that had even marginally acceptable results are #306 and #320.

Dan


Kenneth Harris
 

I feel the winner is par, which is hardly a surprise.  It could stand a slight black point move. The strongest esthetic to my eye are in 302, 306, 311, and 318.  If I judge these in the context of online usage, I'd choose 311.  Strangely, on this particular set, my opinions are affected by which order I view them more than usual.  (FWIW, I'm noticing that Adobe Bridge 18 has problems with ratings/labels done in the more recent versions of Bridge).

Ken Harris


Dan Margulis
 



On Jun 1, 2020, at 12:48 PM, Kenneth Harris <reg@...> wrote:

I feel the winner is par, which is hardly a surprise.  It could stand a slight black point move. The strongest esthetic to my eye are in 302, 306, 311, and 318.  If I judge these in the context of online usage, I'd choose 311.  Strangely, on this particular set, my opinions are affected by which order I view them more than usual.

This is a lot truer than commonly understood. The order can be jiggered so as to get more votes for a certain version. Having seen this effect a lot recently, in these two case studies I numbered them in order of receipt, butI assigned new ones this morning via random-number generator.


Roberto Tartaglione
 

Dear all,
I think the peculiarities of this image are:
-a flaw in the makeup on the right elbow and on the nose between the eyes; this affected my workflow because every increase in saturation and contrast boosted the difference between the skin (with no makeup) and the skin with.
-the veil needed a processing for increasing detail ( without USM, but if it was allowed, it would have been the same) - the face just the opposite.
-the background needed to be "reduced" in order to make simultaneus
contrast with the face.

The difficulties in all this is that we were not asked to play with grading or
other styles ( it would have been easier IMHO), but to fix an image that requested an interpretation.
This is why there are so many differences.

I totally agree with the group, the par version is awesome: perfect pink skin tone, good contrast overall and detail in the veil.

Roberto
roberto@tartaglione.com


bill bane
 

I thought problems arose because it appears that the different sides of the veil had different color lights (flash, and warm/yellow?), and that the face had some combination of uneven makeup and pores. And, contrary to would have thought, veils are always to the warm side of white, except for the very inexpensive ones.

Thus, getting to "correct" color was a complicated dance for me, with some compromises inevitable somewhere. I chose to sacrifice strong face color, toward pinkness, but stayed within PPW skin face ratios. I could have, and should have tried to add face color at the end, but during my first pass though the MMM/CB stage, I found the face colors exploded, and badly, so I only used a small bit (via opacity adjustments) of this at that point. At the end, I suspect I could have been able to add some additional variation (and more pinkness) if I had tried.

That being said, not in the retouching business, I have been around brides and they are a fickle lot. I think the Par version is certainly the best, but I find it "too pink", for what I think most brides would want (actually "would desire", irrespective of accuracy). The Par version remains a bit too snapshot-like and therefore not as "dignified" as I think most brides would want. YMMV, and seem my related comments below.

I also thought the colored flowers and the rear high pipe greatly distracted, or, depending on your perspective, allowed the photo to remain looking too snapshot-like. They are easily desaturated and/or defocused.

Finally, I am not a worthy participant with some or most of you since I have never been professionally involved in anything like this. Therefore, all of what I say could be worse than worthless, but I thought I would stick my toe in the water with an opinion to see if a viper (AKA, Dan) would bite it. 😊

This is also a preamble to say that I may have "cheated." By that I mean I used (on top) the software "Portrait Pro", but only to add facial lighting I know professional portrait photographers create with their fancy setups. In my case it was to add rear and edge face lighting and some facial tonal contrast. I think I could have copied what this program did using normal Photoshop skills, but that would assume I had the experience and taste to know what that lighting should look like. Portrait Pro is very normative, and I presume its normative-ness comes from their user's profession experience. To my taste almost all of it is over the top, but the software allows you to pick and choose among ~100 changes to use. On1 and Luminar also have "portrait" modes, but Portrait Pro is vastly more powerful, controllable, and expensive.

Again, with these effects (and skin colors), I think one is likely to "get it wrong" if one does not have a sense of what the bride's wishes are. Some would regard these studio lighting effects a too "presumptions" or too dramatic, and/or incongruous and ill-fitting with a scene containing orange pipes, orange flowers, and a black pipe handrail. But......to the contrary, some brides, I can imagine, would squeal "more, more!".

Bill

-----Original Message-----
From: colortheory@groups.io <colortheory@groups.io> On Behalf Of Roberto Tartaglione
Sent: Monday, June 01, 2020 1:03 PM
To: colortheory@groups.io
Subject: Re: [colortheory] Veiled Bride: Results

Dear all,
I think the peculiarities of this image are:
-a flaw in the makeup on the right elbow and on the nose between the eyes; this affected my workflow because every increase in saturation and contrast boosted the difference between the skin (with no makeup) and the skin with.
-the veil needed a processing for increasing detail ( without USM, but if it was allowed, it would have been the same) - the face just the opposite.
-the background needed to be "reduced" in order to make simultaneus contrast with the face.

The difficulties in all this is that we were not asked to play with grading or other styles ( it would have been easier IMHO), but to fix an image that requested an interpretation.
This is why there are so many differences.

I totally agree with the group, the par version is awesome: perfect pink skin tone, good contrast overall and detail in the veil.

Roberto
roberto@tartaglione.com


Robert Wheeler
 

Take my comments as originating from someone without extensive expertise in color adjustment. Now that I have retired, I am finally finding time to learn more through reading and practice. I find it challenging to make serial rank-ordering of all the entries, but they do seem to group up into several levels. I’ll list my subjective groupings in numeric order within each group for easier reference. My opinion for many of these changes on repeated viewing, so nobody should ascribe excessive weight to the comments or groupings.

 

My top level group would be 302 (good whites, background subdued tone), 304 (forehead has a little glare, but face is good), 306 (a little on the cool side, but face again nice), 307 (my top individual pick), 310 (maybe could use a bit more contrast), 313, 314, and. 323 (PAR version, looks best to me),

 

My middle group would be 301 (reddish color cast noted), 311(face somewhat orange but eye has good color), 320 (bride good, background perhaps too distinct), 321 (veil seems a bit cool toned), 322 (forehead wrinkles a bit prominent).

 

My lower level group would be305 (seems to have a pale haze over everything, perhaps needs some contrast) 308 (wrinkles somewhat pronounced, green plant color not very realistic, camera left part of veil not seeming white enough), 316 (rather dull and hazy, with haze removal over eye making it look too sharp compared to rest of face), 317 (brick wall replaces all of background but uneven lines make a vertigo sensation, face somewhat dark, cheeks and upper forehead not fitting naturally with adjacent facial tones), 318 (overall a bit dark, veil not appearing very white, but removal of railing helpful),

 

My lowest level group would be 312 (mesh part of veil not distinct, face has strong blue cast and blotchy coloration, eye too bright for rest of face, teeth and lips too bright for rest of face), 319 (face orange tinged with wrinkles exaggerated, background flowers and greenery unrealistically vivid), 315 (blue tones in skin, skin perhaps over smoothed, eye brightness excessive for rest of face, removal of handrail and post ok)), 309 (strong orange in face, background bright, veil not seeming white).

 
Looking forward to reading other opinions and hearing more from Dan.

Robert Wheeler


Harvey Nagai
 


Many thanks to Dan for conducting this case study and to all those who
participated.

Faces and skin tones have always been the most vexing aspects of photography
and image editing for me.

What made this case study especially difficult (and vexing) is that my opinion
of a portrait image changed each time I saw it, sometimes while I was looking
at it.  And as previously remarked, sometimes with what I compare it to or how
I compared it.

But if we can't have a little fun with our frustrations, why bother?

My own version ended up in left field, intentionally.  The usual rules of
thumb weren't working for me, so nuts to b>a and onward to try to make pink
and soft work.  If it didn't get judged well, no worries.

In trying to pick out my favorites, the first question was: would I want
that portrait sitting on my desk staring back at me?

Five entries were eliminated on their own traits.  Six more were eliminated in
comparing them to others.

Comparing the remaining 11 versions against each other and (mistakenly) the par
version led to my final five, including my own, which I eliminated, and par-blended
the remaining 4 to get... something that looked very similar to my own but with
colors others apparently like.

What fun.

HOWEVER, par-blending 5 versions each with many aspects I quite liked resulted
in something I liked more than any individual image, athough it can't hold a
candle to Dan's par version.

Speaking of which, I'm really interested in why Dan chose his finalists (not
necessarily identifying them).  I would guess he didn't necessarily pick
candidates which converged towards a hard ideal to end up with a par blend
not much different than any of them, like I did.


jwlimages@...
 

I will stick my neck out - (I've been a "lurker" here, and have learned some valuable things over the years) and offer a couple comments. I agree this image is surprisingly tricky. I think the veil serves to complicate/confuse judgment of how to treat her skin tones, for color, overall values & contrast range, especially since the default settings on the .dng I downloaded were surprisingly dark (minus 1 full stop exposure), as were the rendered jpgs, especially the one marked "flat". I agree with Gerald that the background needs some help to recede more - I see a few tried darkening and/or desaturating and/or blurring the wall, and I wish I had devoted some time to that on my entry.

 

It's amazing how the par version is again pretty much the winner. My only reservations are that it still looks a bit on the dark side, and the poor girl could use some retouching on her lips, skin blemishes, creases & wrinkles. Also her eye sockets look almost sunken - lightening them up as in #303 helps (even if it looks like a recipe from Portrait Pro on On1 Portrait, or could have been selectively masked to limit it to the eye areas?). I was feeling pretty good about my submission until I saw much better skin color on a couple others, again "perfected" in the par version, plus some other things like the foreground/background distinction that I didn't think to work on.

 

I did try to not put too much time into mine, imagining the poor wedding retoucher with hundreds of these to do. That was a nice change from my "day job" assignments, where they would have had me start by completely 'removing' the part of the veil obscuring her eye. I confess I spent a bit of time thinking about & trying a couple techniques to start that before I caught myself & reread the instructions - "just make it look good" - whew!

 

So thank you for a good learning experience. I look forward to seeing more comments and Dan's analysis.

 

John Lund


Dan Margulis
 



On Jun 1, 2020, at 11:10 PM, Harvey Nagai via groups.io <hnagai@...> wrote:

Speaking of which, I'm really interested in why Dan chose his finalists (not 
necessarily identifying them).  I would guess he didn't necessarily pick 
candidates which converged towards a hard ideal to end up with a par blend 
not much different than any of them, like I did.

I’m working on a post to discuss that, but I’d first like to check on one thing that reponses haven’t mentioned, in fact one response seems to take an opposite point of view from mine.

At the outset of this contest I suggested that the problem was that the veil covered half the face. As at least two people pointed out in their private messages to me, that’s really not the big issue. The lack of color is. The face dominates the image and we are severely limited as too how red we can make it. The background is even duller. The flowers at left can and probably should be intensified but are too tiny to make much of a difference.

We are so desperate for color, IMHO, that the minute we open the file we should see that the eyes need to be made lighter and bluer. This is by no means the only challenge, but it’s critical to success. Flipping quickly through the files and ignoring the ones with obvious other problems, I say that #s 303, 307, 308, 310, 311, 314, 317, and 321 are handling the eyes well. #s 302, 304, 306, 318, and 320 are otherwise acceptable but have overly dark eyes that make the result seriously less attractive.

Am I crazy to put such emphasis on the eyes in this colorless capture?

Dan


ROBIN MARK D'ROZARIO
 

No. The eyes probably make or break this image.
While working on this my approach concentrated on 
1. The neutrality and lightness of the veil at the expense of some detail. It separates the subject from the wall and outlines the face.
2. The face - not too much contrast and no crazy colour. I ended up colder than par though.
3. Getting the eyes improved. If they are poorly done the first two won't be enough. Still while I feel I did a good job on the eyes the surrounding areas above and below ended up darker than par.
I feel the background is secondary as long as it's not distracting from the subject.
I also have a question which might seem impertinent so apologies if anyone is offended by it. I am not trying to start a controversy.
Last project Roberto was the consumer / client and could decide  on what he liked. There is no specific client this time but my opinion is this type of image is for brides/ prospective brides. If used in a catalogue then mainly to be seen by female customers. So far we have been analysing this image on technical merits and all respondents have been male. I ran the images by my wife on a like / don't like, looks good/ doesn't look good basis. The par image got a lukewarm response. It was not rejected but neither was it singled out as being superior.
Dan, since this image has been used before in your ACT classes have you noticed different reactions based on gender?
Best regards,
Robin Mark D'Rozario


Robert Wheeler
 

Here is some additional input from a non-male source. My wife also looked at the veiled bride submitted image set. She does not engage in photography, but does quite a bit of color work making unique and artistic greeting cards that involve colored inks for stamped designs, colored papers, and various paints. She pointed to the PAR picture as the best in here opintion, but commented that it could have been a little brighter. She did not like the images with strong blue color cast, and also did not like those that appeared "over processed" (pressed for specificity: skin with plastic smoothness; background extremely different than subject with excess blur, very different tone, or fake looking color saturation of flowers and greenery; eyes or teeth much brighter than rest of face).

Robert Wheeler


ROBIN MARK D'ROZARIO
 

Thanks. My wife has no background in photography or colour processing. Your wife already has some technical knowledge to help her discern a better product on technical merits.
My observation of my wife's choices  is that  she would generally choose contrast over subtlety in colour or smoother transitions.
The par version is subtle beast. On first viewing I was underwhelmed. After trying several different blends with my submission and some unfinished versions there is always an improvement.

Robin Mark D'Rozario


Dan Margulis
 



On Jun 2, 2020, at 10:47 AM, ROBIN MARK D'ROZARIO <rdrozario@...> wrote:

Dan, since this image has been used before in your ACT classes have you noticed different reactions based on gender?

Not this specific case but I’ve shown hundreds of portraits of women to various clients. Female reaction to portraits of other humans of either gender is different from the reaction of males but it is not that big a deal in the overall scheme of things.

When viewing a portrait of themselves, however, the reaction by gender is enormous. The female reaction is so unpredictable, and occasionally so bizarre, that I make it a practice, when possible, to show women important images of themselves when someone else is present. And I instruct the subject to keep quiet for several minutes while the other person and I discuss the merits of the various options. Present either #303 or #311 to the actual subject, and some will like it and others will literally take it as a personal insult. And a couple of other versions that I won’t name would draw a reaction I’ve seen many times: “This makes me look like a hooker.”

I have never dated a woman who did not believe that she had some horrible physical imperfection that rendered her unattractive to anyone smarter than me. Here the bride is no teenager, but we can give her the skin of a teenager if she wants. Some would, more would not. It may be a matter of politics, or of self-esteem, or of a ridiculous delusion that her skin looks like a crocodile’s.

Tastes also change. In the Chevreul book I point out that every fifty years or so, fashions favor presenting women as absurdly delicate and fragile. I cited Scarlett O’Hara, who had to corset her waist down to 43 cm, and had to shield herself from the sun to preserve her paleness. And the hobble skirts of the early 20th century, and the emaciated “Twiggy” look of the late 20th. As for our century, many of my books feature portraits of women from a professional “boudoir” photographer in Canada. His clients order up glamorous photos of themselves, sometimes for their own enjoyment, sometimes to give to a significant other. Looking over his portfolios over the years, we both marvel at how 25 years ago, the shots, though explicit, were what one would call “tasteful nudes.” In this century, the women are generally asking to pose for what I would describe as pornography.

The short answer, then, is that women other than the subject would probably have more or less the same preferences men do in evaluating our case study. But the subject herself? That’s anybody’s guess.

Besides, last I checked, a marriage requires two parties. The spouse may be under the delusion that he is the client, too, and that he gets a say in the matter.

He’ll soon learn. As Miss Adelaide advises in Guys and Dolls,

Marry the man today,
Give him the girlish laughter,
Give him your hand today,
And save the fist for after.

Dan


ROBIN MARK D'ROZARIO
 

Thanks for your reply..
Thankfully no fist for me yet though a verbal "fist" may make an appearance soon. Carnival was thought of as a one off but the appearance of " raincoat girl" so soon after the bride has been noted!


Harvey Nagai
 

On Tue, Jun 2, 2020 at 08:47 AM, Dan Margulis wrote:
Am I crazy to put such emphasis on the eyes in this colorless capture?
 

In one word: no.

It may be that there is less portrait experience among the partcipants
than in a typical ACT class or in the group list several years ago?

Personally, eye color doesn't matter one bit (says the guy whose family
and relatives all have brown eyes).

The darkness around the eyes I did recognize early on as a potential
problem, but only as something to avoid worsening into "raccoon eyes". 
My reasoning was that she has naturally deep-set eyes, that's who she
is and I shouldn't get hung up on it.

Wrong.

The first thing that struck me about the par version is how open and
bright the face is without looking flat.  It's an eye opener in more
ways than one (yes, I went there).

I'm less enamored about the blue eye in a sea of yellows, it does draw
attention, a bit too much for me, I keep coming back to it instead of
taking in the face as a whole.  But that just might be a minor matter
of tweaking it a teensy less blue.


James Gray
 

Dan,
This question may allow some readers to guess which result is mine.  I do not mind.  I know I am not that good with portraits.
Do you have any idea why the skin on my version looks splotchy when I view it on the color theory web site with Chrome.  When I open the image I sent to you in Photoshop, the skin looks much better.  Is there a way for me to detect this problem while in Photoshop so I do not do it again?

The discussion so far is very interesting.

James Gray



Robert Wheeler
 

James,
It may be presumptuous for me to respond ahead of Dan, but here are some thoughts. In Photoshop, the "View" menu has an item for "Proof setup." This is supposed to allow Photoshop to provide a preview of what your image will look like if output to various environments. For web display, a typical choice for proof color would be "Internet Standard RGB (sRGB)." It is also possible to select proof colors for various printer -paper profiles. The View menu has another item, "Proof Colors" that toggles your view from whatever you have selected to work in (under the Edit menu "color settings" dialog) to the proof colors you select in the View menu. Looking at your image using proof color settings may provide a better approximation of what the image will look like outside of Photoshop.

Of course, there is more to the story, as I have been finding out after recently moving to a new monitor with a new (to me) color management implementation and a new colorimeter. How you color-calibrate your monitor may impact how things look in various programs. Some calibrators set up ICC profiles that can be selected from within different software programs. Other calibrators embed color lookup tables (LUTs) into the monitor hardware (and the operating system needs to have the right settings to work with the LUT). My understanding is that Windows handles color with different settings and profile locations than the Macintosh operating system(s). Many programs default to using the color translation method designated for the operating system to work with the monitor, but other programs allow the user to specify specific profiles. Setups with more than one monitor may depend greatly on what capabilities the video card inside the computer provides for handling the second monitor (often only the main monitor is correctly color profiled).

The web browser can be part of the complexity as well. A few months ago I looked at a number of a number of online resources about how various web browsers do or do not interact well with color management in the host computer or with color management indicators embedded in web sites or color profile indicators within images posted on web sites. I don't have specific links handy, but simple search should produce several. I mostly use Firefox, so am not expert in Chrome color management settings.


Jim Sanderson
 

Thanks, what color problems did you observe on Firefox?

, with purpose and great thought. 

On Jun 2, 2020, at 5:24 PM, Robert Wheeler <bwheeler350@...> wrote:

James,
It may be presumptuous for me to respond ahead of Dan, but here are some thoughts. In Photoshop, the "View" menu has an item for "Proof setup." This is supposed to allow Photoshop to provide a preview of what your image will look like if output to various environments. For web display, a typical choice for proof color would be "Internet Standard RGB (sRGB)." It is also possible to select proof colors for various printer -paper profiles. The View menu has another item, "Proof Colors" that toggles your view from whatever you have selected to work in (under the Edit menu "color settings" dialog) to the proof colors you select in the View menu. Looking at your image using proof color settings may provide a better approximation of what the image will look like outside of Photoshop.

Of course, there is more to the story, as I have been finding out after recently moving to a new monitor with a new (to me) color management implementation and a new colorimeter. How you color-calibrate your monitor may impact how things look in various programs. Some calibrators set up ICC profiles that can be selected from within different software programs. Other calibrators embed color lookup tables (LUTs) into the monitor hardware (and the operating system needs to have the right settings to work with the LUT). My understanding is that Windows handles color with different settings and profile locations than the Macintosh operating system(s). Many programs default to using the color translation method designated for the operating system to work with the monitor, but other programs allow the user to specify specific profiles. Setups with more than one monitor may depend greatly on what capabilities the video card inside the computer provides for handling the second monitor (often only the main monitor is correctly color profiled).

The web browser can be part of the complexity as well. A few months ago I looked at a number of a number of online resources about how various web browsers do or do not interact well with color management in the host computer or with color management indicators embedded in web sites or color profile indicators within images posted on web sites. I don't have specific links handy, but simple search should produce several. I mostly use Firefox, so am not expert in Chrome color management settings.