Red Rose: results
I’ve posted the results of the last in pir series of 11 studies.
Reviewing: This seems like a fairly routine shot of a rose, the sort of flower image found in every photographer’s portfolio. It is from the MIT study and there are no specific instructions on what is the desired result.
We have 21 entrants. Most people 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 #1201 to #1221. As with past studies, we also have a “par” version, #1222. To get it, I chose what I thought looked like the five best entrants, and averaged them, each one weighted 20%. This often creates a version that is superior to most if not all of its parents.
I’ll write about the individual submissions, but as usual I’d like to open it up to group discussion first. Unlike previous case studies, however, I have a couple of words in advance.
This one is very like the Carnival study that started our series, except that one was in CMYK and this one in RGB. In both, the central object is extremely red; we likely want it to appear as saturated as possible consistent with holding detail.
It is correct that this specific problem doesn’t come up that often, but you need to be able to handle it because the techniques that improve the reds of the carnival costume and this rose also work for less saturated reds, which happen to be the most crucial color much of the time. Consider the list of our case studies:
Toast to Greece
Seated in Grass
Several of these have reds, greens, and blues of approximately equal importance. That’s the case in Niagara Spray, Cinque Terre, and Panama 1978. It would be nice to get excellent shape in all these colors, but if one of them isn’t that good, yet isn’t a total disgrace, then it isn’t necessarily disqualifying.
But what about Carnival, Veiled Bride, Monument Valley, Seated in Grass, Adirondacks, and Red Rose? Well, in each of these if you have second-rate control of reds, then your version is a failure, full stop.
These images were chosen for their diversity, trying to get a decent representation of the sorts of work that crosses the screen of professional retouchers. Control of reds is critical in half of them. So I would look closely at how well you did on this rose image because it may be a predictor of results on more important ones. And then you might want to rethink some of the steps you took on the other five red-critical exercises.
The folder is in the group Photos section, named Case Study: Red Rose
I also have zipped all 22 files and uploaded a 16 mb file to our Files section,
Search for Red-rose_entries_080320.zip
If you are going to study these versions I strongly encourage you to download these files. Many of these entrants vary only in a minor way and it is hard to see the impact of a change without toggling back and forth between them.
I look forward to your comments.
I for one appreciated that the subject was something that I have seen, maybe not this exact rose but one that likely is similar. What I wanted to do was to reproduce as much as possible what I thought a rose should look like... maybe becoming a bit romantic and pushing the colors just a bit but not so much as to be unbelievable. Then getting the greens right since my perception is that rose leaves are just a bit deeper than most leaves (so somewhat darker). That said, this was an interesting exercise comparing to the other submissions.
I would guess (probably incorrectly) that 1212 blnd-v1-v2-CB.jpg, and 1203 Rose&Bed.jpg which are darkest in the rose petal ends versus 1205 RoseV4.jpg and 1221 4475-down.jpg which are considerably lighter would result in a good blend for luminosity. Plus a blend to tame the colors using 1207 RR9-a2074.jpg would help some of the supersaturated ones. So with just a few of the others to blend into mine, I felt I could get a big improvement (which I tried and verified).
I am just a lurker, dont’t have time to partake in the challenges - not my day job. But this issue of how to best deal with challenging red flowers is one I am super keen to understand as I often have this problem in my personal photos. Look forward to closely following the discussion on this one and thank those who took the time to contribute and share their solutions.toggle quoted messageShow quoted text
On 3 Aug 2020, at 12:23, Dan Margulis via groups.io <firstname.lastname@example.org> wrote:
I worked from the flat version of the rose. The original has a nice, soft appearing rose with, perhaps, more light to white areas in some of the central petals than ideal but with an overall haze/inadequate contrast. On the left side of the image, white blotches and a dark area without greenery have contrast pulling attention away from the subject without providing any reward. My entry is 1205. I did darken the bright white area, but not enough to actually resolve the problem. After submission, I verified that cloning to replace the white and black with foliage makes a huge incremental improvement (a bit frustrating since that was not allowed). Entries 1211, 1212, 1215, and 1218 turned the white to dark grey (to good effect). 1214 and 1217 may have used cloning to remove the white blotches and show a bit of greenery instead, nicely improving the image (whether allowed or not). The PAR version has the white blotches darkened well.
For color correction, I blended two versions. One used the MMM+CB action with rose selected but without preliminary contrast adjustment. I applied mild tweaks to the endpoint adjust layer, followed by a curves layer to darken the L curve in luminosity mode at 75% opacity. Return to sRGB was followed by curve masked to darken the white areas and make a mild vignette. A final Auto tone layer at 55% completed the job. The rose still appeared like a nice soft flower and had stronger reds than before, however, the central light parts remained too light. The second version started with shadows and highlights action followed by MMM+CB using select all. Return to sRGB was followed by duplicate layer in multiply mode, masked and painted to darkened the white blobs and add a vignette. This version had areas of over-saturated red. Blending the first version with the second at 50% opacity in normal mode produced a gentle rose color and more color centrally.
I find the PAR rose to have good color throughout although the somewhat strong contrast makes the flower look less soft to my eye. Other entries with harsher contrast seem to make the flower less soft appearing to my eye. I did blend one of the darker entries into mine at low opacity, with some improvement to the central color of the flower.
The green foliage shows a shadow from the flower consistent with bright sunlight. I thought submissions that darkened the entire background resulted in a less believable image. Some appeared to destaurate the green leaves, which left the flower as the most colorful object but abandoned the useful red-green color contrast.
As this series winds down, I send my thanks to all the group participants who submitted entries and commentary over these many weeks. I have found it exceptionally valuable to evaluate so many different versions of the same image. Measuring LAB values across sets of images has greatly improved my understanding of LAB. Comparing my own versions to many others has expanded my understanding of how many ways there are to approach image improvement and opened my eyes to what can be done using different technical approaches. All these images provided an exceptionally valuable environment that helped make color correction theory much less abstract while reading and re-reading Dan’s books and the PPW action set documentation.
And I especially thank Dan for organizing the challenges and generously devoting time to comment in detail on so many images. Perception involves more than the eyes. Early on, many of the variations of images in your books looked almost identical to me. Now, the differences are much easier for me to see. In addition to learning advanced post-processing techniques, I have also learned a great deal about how to be more specific when evaluating images, how to be open to a breadth of variations, and how to become aware of my own preconceptions and biases. Having the learning spread out over multiple weeks has been a wonderful luxury, allowing time for experimentation and reading that might not be possible in a more intensive setting. I am deeply grateful for your expertise and help.
toggle quoted messageShow quoted text
I did not find it difficult to improve contrast in the rose using channel blending and by darkening the rose. That killed some of the saturation. The hard part for me was knowing how far to go. I guess I went too far. The par version is superior to mine.
Mine is 1207.
I struggled with how to get more colour into the lighter petals without having it all unrealistically red.
In my final version, I blended four versions were one was quite yellow / orange, one was blue-ish, one with flat colours and one with H-K. In my view, this had the ingredients for a warm colour in both the rose as well as in the greens.
The result is ‘more yellow than blue’.
At the time, I found this to be reasonable, as the picture was shot in sunshine.
I think some of the entrants are very good at getting the colours more saturated and still keep the contrast and the green. Particularly, I find 1201, 1214 and particularly 1216 to stand out. I would like learn how 1216 was done.
I do like my own as well, but see that I fail in getting the redness I hoped for.
Even though this is hard, I find it most interesting and rewarding. After each picture, I try to get the PAR version by following the hints from all you guys, and particularly listen to what Dan says. Some times I get close, and other times I do not.
I greatly concur with the words of Robert Wheeler, as I was barely able to participate in 2 challenges since I was ( surprisingly) busy with online education during these months, and most likely my Red Rose delivery was either rejected or did not arrive on time, and to be perfectly honest, I am just a brand new rookie to Dan's workflows and techniques. I am still reading his latest Photoshop book, along with the Chevreul reading and I need to sit down formally to understand what are Dan's tools doing to the files. I used to be a scientist in my former life, so I can not just punch buttons and see what happens, so I have downloaded all recent challenges and will study them against my own work, as I progress in understanding this totally new approach to Photoshop workflows. I have no single idea about working on LAB, so you can really get it from there... I am still a workflow cripple.toggle quoted messageShow quoted text
In addition, my work is mostly on fashion and people so, I can clearly see that fixing landscapes and flowers is definitely a more technical process to achieve a subjective result. Quite a conundrum!
On a very specific note I need to understand like right now what is the "Convert to RGB" button in Dan's PPW doing, since I saw barely no change to my Red Rose file when I clicked on that button but when I tried to do it on my own, the changes in color were disgustingly unsaturated to say the least, and that was quite something to watch!
That said, I really enjoy seeing so many varied interpretations of one single image that I am amazed! From looks I would never consider doing all the way to images I want to be able to replicate, this challenges will be a very good base for my new learning, so, many thanks to Dan and all participants. Hopefully, I will be more capable and trained to engage in the next challenges, should Dan agrees to do something like this in the future.
PS: On a side note, and acknowledging again that I know nothing about LAB processing, I wonder if some of you guys have tested this small application called RPP or Raw Photo Processor 64, which I downloaded a few years ago ONLY for the strange ability this app has to generate a LAB file in 16 bits directly from a RAW file. As you guys spend a good deal of time in LAB ( I will learn about that) I wonder if this could be a useful tool to some of you, or if you tried it and discarded it??
On Tue, Aug 4, 2020 at 05:05 PM, Robert Wheeler wrote:
In reading over everyone’s descriptions it was striking that some people, although they got good versions, had to struggle mightily to do so, often making and discarding several versions. Meanwhile, the experienced folk, who have seen this kind of image many times, could knock out something competitive to the par version in just a few minutes. They know that the secret to shape in a red object is the red channel itself, and in something this bright the camera usually does a poor job in establishing detail in that channel.
Therefore it’s almost automatic to blend something into the red, usually in Luminosity mode, but there are many alternate ways to get the same result. Without a decent red channel the flower lacks depth. Similarly, without a decent red channel the subject in Seated in the Grass lacks a nose. We saw examples of this in both case studies.
It is also possible to take this method too far and while the result may seem “realistic” it can also seem harsh, lacking in the softness that we prefer in a flower or the face of a young adult.
Putting too much contrast in this particular flower backfired on some people when the lightest parts turned white and the deepest reds almost a black. Even though the rest of their rose is as red as anyone else’s, all this neutrality causes us to perceive it as grayer.
Exactly how red to make the rose, how dark to make it, and whether to move it toward orange: these are personal decisions without right or wrong answers. Two other choices, however, seem to be clearer.
First, the camera picked up some nasty white reflections in the top of the flower. I don’t see how they can be natural. Like other physical defects we’ve seen (noise in the Colosseum and Toast to Greece; the patterning in Panama 1978) it’s a bad idea to try to obliterate it. OTOH we certainly don’t want to make it worse, and some people did that. I’d say the par version shows how to handle it.
Second, and more interesting: like Seated in the Grass, this one features a red foreground object and a green background. In that one many of us, including me, favored making the background an unrealistically bright color, feeling that this set off the woman’s fleshtone with little downside (since there was no real detail in the grass to emphasize our incorrect color.)
From that, one might suppose that we should go with leaves in this rose picture somewhat more vivid than the dull green they are in real life. The results seem to me not to back this theory up. They shouldn’t be made gray, like some people did, but dark and dull is more effective than light and bright, whether or not we choose to put a lot of detail in the leaves.
Why this contradictory result? Chevreul certainly suggests that green, being complementary to red, is not a bad thing if we’re trying to emphasize that something is red. But this intense flower doesn’t need any help, the way a fleshtone does.
Also, in one of my chapters in the Chevreul book, I suggested that the longer it takes for us to recognize the image we’re looking at, the more prone we should be to exaggerate simultaneous contrast. Seated in Grass has several elements, so it takes a few picoseconds to grasp what it’s about. This rose, OTOH, we recognize instantaneously. That’s why, IMHO, the color portion of MMM isn’t particularly effective here, and perhaps explains my preference for duller greens here but brighter ones there.
I’ll have comments on the individual versions later.
ROBIN MARK D'ROZARIO
"I would like learn how 1216 was done."
The short answer is by the seat of the pants/trial and error. I am not a professional photographer or retoucher. My first attempt was made by blending RGB into the red channel. Subsequent channel moves (an area where I still need much improvement) were poor and I ended up with a ghastly contrasty result. I went back to a more rudimentary approach as outlined below.
For the entry.
DNG conversion in DXO Photolab. White balance from white flower, Colour rendition to Generic color neutral Gamma 2.2, Lens sharpness module on default and noise reduction on default. In these exercises I have found these settings for DNG conversions give me a starting point between the default and flat files supplied, though closer to the default in look.
In this case it turned into a pink rose!
1.Curves layer in Luminosity mode.I was trying to open up the flower . End points of the three channels pulled in from right. The red was lowered and pulled in.The green and blue as well. The long tail of the histogram(probably from the white flower) was clipped off, more by the blue than green. Most of the increased contrast comes from the green here. How much the end point of the red is moved down/left or right of 45 degrees will determine the luminosity separation/saturation between the the different shades of red.
*In hindsight I should have moved the red endpoint lower. I feel that in my submission the lighter petals are too light.
2. A Hue/Saturation layer with Red hue moved 21 points to the right. Luminosity mode. Adds more separation between light and dark reds.
*Post submission I revisited this step and the benefits are greater in colour mode, a darker richer red post H-K which I believe would have gotten me a better overall result. I was too focussed on Luminosity earlier.
3.A selective colour layer increasing C,M,Y in red. I also
reduced Black in Magenta (unnecessary and needing correction two steps later).
4.A Hue/Saturation layer self masking to lighter petals shifting the hue towards red.
5.I now had to correct for the magenta move in 3 with a pulled down luminosity curve to the rose to reduce luminosity in lighter petals. Would have not been needed if I had also got the red curve right in Point 1.
6.H-K through a Saturation mask. The outer petals get more desaturated than the lighter less saturated petals and background.
1.MMM via select all at default settings.
2.CB a>b at 20% using an inverted a channel to limit effect on rose.
3.Endpoint brought in to edge of histogram and slight s-curve.
4.A Hue Saturation layer to slightly desaturate shadows on rose petals .
5.Blurring, darkening and painting with green on white rose in background.
6.Lens blur filter to left and top background.
7.High pass in overlay mode 2 pixel radius for fine detail.
8.High pass in overlay mode 16 pixel for shape.
9.Shadows were plugged up so I made another Lab version with S/H before MMM limiting effect to the darker shadows. All other steps were similar.The rose was very similar in both versions. Blended both versions at 50% .
While I am happy with the shape of the flower I think the overall redness should have been more in line with par, 1209 or 1218. The lighter petals still retain a little too much pinkness. Much of this could have been avoided if I had been more careful with the red channel luminosity curve in my first move.
Hi to all! Mine is 1218. The Ferrari red one!
I wanted to make it as R as I could and as "soft" looking as possible. This softness came with the price of not being as detailed as the PAR version. I blended mine to the par at 53% and the result has better detail than in my original.
From the start, the bluish highlights bothered me so I "fixed" them by making them more tonally like the rest of the rose and that is what I found the hardest to accomplish.
All the best!
toggle quoted messageShow quoted text
My entry is 1213.
I was a bit lost in the wilderness as to the color of this flower as well as the contrast when I started.
I spent some time paging through Dan’s books to get an idea of how to regain the contrast for the petals. I finally decided on blending the green channel into the L channel at about 50%. This gave me plenty of contrast!
I tried a couple of corrections from the flat jpg, but then went to the DNG. What bothered me about the image was the blown out highlights, as I saw it. Also, it seemed that was from too harsh of a flash, as determined by the white petal highlights and the high contrast shadow. Therefore, I changed the color balance to flash (5500 K) and decreased the exposure two stops.
I was determined to get rid of the white on the rose, and I still managed to get a luminosity range in the flower, but I can see that compared to everyone else, I might have tried to put back those 2 stops of exposure. I guess, Dan, that I fell into the "obliterate it" group.
It is a great learning experience for me, to explore the different ways to get the color and texture. But as Dan noted with other people’s comments, it was a struggle. I suppose my lack of experience has much to with that. As Dan also said, those with experience (and expertise) could knock out a competitive version in a few minutes. I also spent time going down some dead end pathways.
Once I saw what the par version looked like, I was able to reproduce something close to it’s color in about 10 minutes, without the angst of public presentation. As Dan mentioned in his introduction to this final opponent, if you aren’t having success with reds, look to change your workflow. What made this more challenging to me was not knowing the color, and thinking the original was off, as I said above. The first entry, Carnival, was challenging, but there was a specific, if unattainable, color to reference.
I have been taking another look at my corrected photos, especially those hanging on the walls around the house, and there are a few I am going to redo! (My wife’s face is equal parts a and b. Plenty of red to work with) That said, my rose printed out nicely on Legacy Platine. I hope to print a better result tomorrow.
Robert, thanks for sharing your technique. I wound up blending two intermediate images, and also used a false profile multiplication technique.
Paco, I enjoyed the texture you accomplished in the interior central petals. Very lifelike. I had a problem as well with blue/purple tint near the end of my process. I wound up using a color fill layer with B=88 and painted over the bluish ares to get a positive b value in LAB.
I would echo Robert’s compliments to the group and especially Dan.
To close, I wanted to share a few things:
What I learned this summer!
For the individual comments that I will post in the near future I did some research that turned out not as helpful as I hoped, but it can help explain to some folk why are dissatisfied with their own version. But it also gives us an easy way to get an idea of what the correct procedure for the background.
For each image (and I’ll show the results in the later post) I selected the entire flower and averaged it. That is, not just a single point but the entire rose.
##1209, 1210, and 1214, which are all good versions,happen to have damn near identical average values for the flower. The difference? One person deliberately made the background leaves lighter and brighter. A second felt that they should be made darker and duller instead. And the third did neither and came down in the middle between the two.
If you compare these three versions closely, therefore, the big factor in which one(s) you like is going to be the treatment of the background, since the three flowers are so similar.