Dynamic range improvement

rutt

This happens to me a lot. I have a shot that works but has some blown highlight details. If the shot was taken raw, sometimes it's possible to recover the details with a second raw conversion. But then what? I've had mixed success taking the next step. I have two images, one a success story and one has been an source of considerable frustration.

Let's save the success and start with this image:



Nice enough, but the blown background could be so much better. The background was beautiful. Here is a shot of the window without the girl taken from outside with the background properly exposed:



The girl in the window was shot in raw, so a second conversion was able to recover the background:



OK, so how to combind the two images of the girl in the window?

The most obvious approach is to make some sort of selection. For example, I select the darkest areas of the second image, clear them and then layer it on top of the origiial. In theory this sounds good, but in my experience:

1. It always looks fake, even when done very professionally. I can tell right away that the Singapore girl was superimposed on the harbor in those magazine ads.
2. It's never as easy as it seems it should be. In this case there is a wisp of hair over the girl's forehead that is very hard to select correctly.

I've tried really hard to learn to make these selections and make them work, but I guess I have more hope for something that works by color theory.

A second thing I've tried is various software that can combine over and under exposed images to result in improved dynamic range. I've tried a few of these. One is Fred Miranda's Dynamic Range Improvement Pro. Here is what it did with the two versions of the girl:



Better, but no cigar. I have to admit that it's better than I've been able to do. But I want the background and the girl and I want them to look natural together. If possible I don't want to have to slave to make a selection. So far, I have hated all my efforts.

I suppose that the real lesson here is that I should have used a flash when I shot this to equalize the light from the foreground and background. Live and learn. But there is no next time for this particular image. So I'd appreciate any help. All the full resolutions of the image are here. The raw image is here.

OK that was frustrating. Here is a success story. I started out with this:



and ended up with this:



It's subtle (compared with the improved fall colors), but I recovered the clouds and now there is a sky instead of a white blank. Here is how I did this:

1. Image->Mode->LAB
2. Duplicate layer, blending mode multiply:

3. Apply the following curve to the L channel of the uppermost layer:

After this move, the topmost layer looks like this:



And the blend looks like this:



From here my conventional technique worked to get to the final version: A+B steepening. L sharpening. Move to CMYK and establish white and black points. Fine.

What's going on here? I didn't even need a raw conversion. My curve was extremely steep in the highlights and this brought out the hidden detail in the sky. But this left no room in my gamut for the fall colors and trees. They would have to be either way too dark or else have lost their detail. So I flattened out this curve to keep the non-sky areas as light as possible. That way the multiply blended the good sky with the white sky to get the good sky (since white is the identity element for photoshop multiply blends.) By keeping the rest of the shot as close to white as possible, I had minimum impact on the non sky parts of the image, nothing I couldn't fix with curves afterward.

Here the problem I fixed wasn't nearly as bad as the problem with the first picture. Perhaps that's why I succeeded this time.

Anyone still with me? Anyone think this is an interesting game? You don't have to use my image. I'd love to hear success and failure stores with other images. Let's play.

tmlphoto

John, I don't really know a fraction of what you do about photoshop, but I thought my naive approach might tickle some brain cells or inspire some new thoughts about how to solve this problem. I agree that masking is difficult and is really hard to do without looking fake. While this certainly isn't the final answer I was able to improve it a little. Recipe to follow.


Recipe:
Place the darker picture as the background layer and the lighter picture on top. Fill set to 50%.
Used Shadow/Highlight to further lighten shadows. Used 100% with 100%tonal range. (seems extreame)
Used curved to lighten the midtones and darken the highlights a little more.
Cloned out dust bunny in sky .
Increased saturation on the blue channel only.

Perhaps by using two versions that were slightly less extreame that something like this might work. Also some more advanced blending technique might work ( I really don't know much at all about the different blending modes, but I know it is something I should learn.) I did all this on the M.jpg. If I get some time I'll play around with the RAW file.

One more try. This is the dark imaged with the midtones pulled down in curves layer stacked with my final image from above at 50% fill with H/S at 50%. Starting to look a little funky, but the sky is better. I think maybe stacking the right combination of images from the RAW file could produce a satisfactory image, but it would require a bit of trial and error. I think I have demonstrated both trial and error.

cletus

Great Posts Guys!

ian408

Before:





After:



It is a slightly different starting image but to get the clouds back? Rutt,
you da MAN!

Thanks,
Ian

wxwax

Rutt, why not just use a Layer Mask for your two exposures? Erase the stuff you don't want. Or it that the kind of task you're trying to avoid?

rutt

wxwax wrote:

Rutt, why not just use a Layer Mask for your two exposures? Erase the stuff you don't want. Or it that the kind of task you're trying to avoid?

Exactly the kind of task I want to avoid. I'm terrible at it and the results nearly always look unnatural to me. I've found that it can work with blurring the background sometimes. But the very fact that the background is being blurred and that the exposure of the background isn't changed makes it a lot easier.

But I'm willing to learn. That's why I posted this. Please show me that I'm wrong.

tmlphoto

OK. Here is another attempt. Better, I think. Not perfect, there is some artifact present at 100% mag, but they seem minimal at normal viewing size.
Recipe:
Used the "extract" filter to extract the girl and building and then pasted them onto the sky background. Used a little cloning at some of the edges, but not too much hand work. I used Scott Kelby's extraction method. What do you think John?

mercphoto

The cure

rutt wrote:

This happens to me a lot. I have a shot that works but has some blown highlight details. If the shot was taken raw, sometimes it's possible to recover the details with a second raw conversion. But then what? I've had mixed success taking the next step. I have two images, one a success story and one has been an source of considerable frustration.

Let's save the success and start with this image:

Use your flash.

rutt

tmlphoto wrote:

Used the "extract" filter to extract the girl and building and then pasted them onto the sky background. Used a little cloning at some of the edges, but not too much hand work. I used Scott Kelby's extraction method. What do you think John?

Well at normal viewing size it does look good, but as you say there are artifacts and probably too much error prone work. See the artifact under the shorts? What is Scott Kelby's extraction methond?

I'd like something that works well enough for really large prints, and in my experience, selections and local moves just don't. Often it can look good enough for a dgrin or FM post or enven contest. But when it comes to a 13x19 print, ithe edges are obvious.

rutt

mercphoto wrote:

Use your flash.

Yes, I think that's the big lesson of this particular shot. I even said that in the initial post. But that won't save this image and I can't reshoot it (trust me.)

But there are many reasons not to give up on this topic. The flash wouldn't have helped the fall landscape. If there were glass on the window, it might cause reflections. I tried using multiple flashes to shoot a indoor panorama in order to equalize the light with the outdoors, but it didn't work very well. The rroom was just too large and the light from the windows too bright. It was no match for my three flashes. I think I would have needed some kind of heavy duty studio lights and that was just outside the scope of the project. I ended up using some sort of selections fo that project, but was never really happy with the result. You can see the reslts here.)

Sometimes you just can't use a flash at all because it's too intrusive or because you didn't bring it (I know that last is lame, but there it is.)

So the flash lesson is important and I won't forget it, but it doesn't change the importance of this topic.

tmlphoto

rutt wrote:

Well at normal viewing size it does look good, but as you say there are artifacts and probably too much error prone work. See the artifact under the shorts? What is Scott Kelby's extraction methond?

I'd like something that works well enough for really large prints, and in my experience, selections and local moves just don't. Often it can look good enough for a dgrin or FM post or enven contest. But when it comes to a 13x19 print, ithe edges are obvious.

Scott Kelby Extraction Method
1- Filter/extract
2- Trace image with "edge highlighter" tool. (fairly quick to do, doesn't have to be perfect)
3- Highlight part to extract
4- Extract
5- Duplicate the layer, use History brush to fix "dropout" areas
6- Choose background picture, paste onto new layer of extraction pic.
7- Put background layer behind extraction.
8- Use erase tool to clean up

I think it would be very difficult to find a method that would work at the larger print sizes without a fair amount of manual work, but this method seems to give a good start. Admitedly, this was a fairly easy extraction. It did do quite well with the hair though. A little fill flash would go a long way. It might be nice to use the flash with some negative flash compensation so that the effect is more subtle and only partially brings up the shadows and then use Shadow/Hightlight to bring them up a little more. Well, I think I have reached my limit on this one.

rutt

Thanks, Thomas. You did a lot. I'm still hoping for a global technique like the one I used for the sky in the autumn picture. But I'll have to try your methods out when I get a chance.

wxwax

rutt wrote:

Exactly the kind of task I want to avoid. I'm terrible at it and the results nearly always look unnatural to me. I've found that it can work with blurring the background sometimes. But the very fact that the background is being blurred and that the exposure of the background isn't changed makes it a lot easier.

But I'm willing to learn. That's why I posted this. Please show me that I'm wrong.

It is time consuming, no question. IMHO it will yield the best result in this extreme example. I'll do it this evening when I get home, you can see what you think. Masking has the added benefit of allowing you to darken the walls slightly, so that the lighting is a little more balanced.

tmlphoto

You have tapped into my compulsive side. I ran a small "blur" brush around the extraction edge and fixed a couple of artifacts. Still not perfect, but I think with a little more time and attention this method could work well. No pain, no gain .

rutt

wxwax wrote:

It is time consuming, no question. IMHO it will yield the best result in this extreme example. I'll do it this evening when I get home, you can see what you think. Masking has the added benefit of allowing you to darken the walls slightly, so that the lighting is a little more balanced.

Thanks. If you do spend your time doing this, please work with the full sized images and save a high quality version (jpeg2k lossless, tiff, psd, jpeg @ quality 12) I like this shot a lot and have been pretty frustrated by it. I'd love to have a version that works. Thanks again.

miketaylor01

PS CS may be the answer...
Another possibility. I know none of you want to hear something good about the Sigma DSLR's but I thought of a possibility on how you could slove this problem. Sigma Photo Pro is the program that is used to read the raw files out of the SD9 and SD10. The are ways to do this with PS CS but SPP has some very interesting features that make fixing problems like this very very easy. They have a feature called Fill Light that is used by moving a slider to add light to the underexposed portions of a photo without affecting the properly exposed or overexposed regions of the image. Dont ask me how this works because I dont have a clue and noone on the Sigma Forum at dpreview has a clue either. All we know is that it works wonders. Here is an example

Before

After

This took about 5 seconds to do. There is aot more I could do to this image to get rid of some of the obvious deficiencies, I just wanted to point out how the fill light feature works and thats the only adjustments I made to this image.

I do not have PS CS, it is in the mail on the way to me , but I have been told that it has a new feature called shadow/highlight that is very similair to the Sigma fill light feature. I assume that it would not work as well on jpg's as it would on a raw file so working on the raw file in CS would probably be the answer. I know that 90% of the time I can recover the underexposed portions of my images when I underexpose the original intentionally to retrieve the blown highlights. SPP makes this very easy to where most of the time I am automatically underexposing the image by around 1 stop and adding 1 stop of fill light. I wish I could Just figure out how to do this in CS for you and then post how it is done, but I dont have CS to try it out yet. Mybe in a week or two I will have a workflow posted on here for you expliaing how its done. I just thought I might tell you that alot of people who own a Sigma say that this feature works close to the same way that the fill light feature in SPP works. And it almost always work for me. Give it a try if you have CS. Or buy a Sigma. lmao just kidding. I bet some of you were expecting that at some point in here. Anyways, I think shadow.highlights in CS might be your answer. Let me knwo if you try it out and how it works.

mercphoto

Oops

rutt wrote:

Yes, I think that's the big lesson of this particular shot. I even said that in the initial post.

Mea culpa, I missed that.

But there are many reasons not to give up on this topic.

Agreed. But alas, its way over my head to contribute, I'll just have to read and see if I can understand any of it.

wxwax

rutt wrote:

Thanks. If you do spend your time doing this, please work with the full sized images and save a high quality version (jpeg2k lossless, tiff, psd, jpeg @ quality 12) I like this shot a lot and have been pretty frustrated by it. I'd love to have a version that works. Thanks again.

Cool, that's what I've been doing, I did a wee bit this morning.

hutchman

tmlphoto wrote:

You have tapped into my compulsive side. I ran a small "blur" brush around the extraction edge and fixed a couple of artifacts. Still not perfect, but I think with a little more time and attention this method could work well. No pain, no gain .

Another way to do this is to select the extraction and under the "select" menu, select "Feather" (or Alt + Ctrl + D for PCs). I usually use 2 or 3 pixels for the feather. Any more and the selection has a strange look. This works pretty good for most extractions and is faster than using the blur tool.

IMHO!

Hutch

Sam

Digital Blending
I have only used this once, but it seemed to to a good job.

Sam



Digital Blending

ginger_55

Sam wrote:

I have only used this once, but it seemed to to a good job.

Sam


Digital Blending

Thanks Sam, I just printed that stuff out. I have a method I got out of a book, but when I did this assignment, I found my method to be a pain, and the results were not usually worth the trouble.

g

wxwax

I haven't forgotten, rutt, still chugging around her legs.

arjunrc

Starting smart
Hi,

There are a lot of ways to enhance dynamic range as posters have commented above. The trick is to select a method that brings you closest/closer to what you want to achieve and then manually masking away what you dont need.

Specific to your picture, it took me less than a minute to do this:

NOTE: Ive only done the automatic stuff - once you reach this far, considering using a low opacity brush around the subject to get rid of the subtle halos



I did the same thing as the 'contrast masking' explained by someone else,
but reversed the process somewhat.

The exact steps:

1. open both light and dark images
2. copy the light image and paste it over the dark image (opposite of the typical contrast masking process)
3. go to the dark image layer, copy it (ctrl-A + ctrl-C)
4. go to the light image layer, create a layer mask, alt-click on layer mask
5. paste copied image onto this layer mask (should show up in grey)
6. gaussian blur the _layer_mask_ by 11 or 12
7. Invert the layer mask (ctrl+I)
8. click on top image layer (till now you were viewing the mask)
9. just a quick 'S' curve for a pop. ta-dah

Suggested improvements (that I did not do)
* use a light brush around the subject to better blend her in

The logic of 'why I selected the opposite layering' as compared to typical contrast masking:

* The only interest point of the dark image is the sky. Everything else is too dark to use - infact its almost black. Aha ! thats perfect. That means that the "dark" image is the perfect mask.

* The light image has all the right exposures but the sky is all messed up. Aha ! thats perfect, since my dark image is so dark, that only the sky is useful.

Therefore, pasting the dark image into the layer mask of the light image does this - it creates an almost black and white mask where the sky is all white everything else is almost white. Wait ! but I want the sky to be black (ie masked) - so I invert it.

The gaussian blur just ensures a smoother transition.

So, you can see, that had you decided to do it the other traditional way (dark over light, what you would land up doing is creating a 'light' mask over a 'dark' image which is so dark that the mask is not very effective.

Creating a 'dark' mask over the 'light' image seemed more logical here

Moral 1: Staring smart avoids a lot of labor.

Moral 2: Spend effort relative to what you want to do with the image. If you are planning on a large print, by all means, blow up the magnification and work hard to iron all edges. If you want to stop at 4x6 a lot of it will not even be noticed.

regds
arjun

wxwax

Excellent stuff, arjun! I'll give this a try at home this evening.

wxwax

Very educational, arjun. I've never done any of that before. The one thing I was not able to do, was make the outdoors darker. Perhaps if I had better Curves control. I must remember this, and try it again.

I'll keep slogging along in the time consuming mode of doing a regular mask, see how the finished product looks. It's OK so far, but as has been pointed out, it sure is time consuming.

wxwax

arjunrc

wxwax wrote:

Very educational, arjun. I've never done any of that before. The one thing I was not able to do, was make the outdoors darker. Perhaps if I had better Curves control. I must remember this, and try it again.

I'll keep slogging along in the time consuming mode of doing a regular mask, see how the finished product looks. It's OK so far, but as has been pointed out, it sure is time consuming.

Hi Sid,
glad this was of help to you. Your image looks great. By outdoors being darker, do you mean the walls ? In this image it should be very easy since the walls have very straight borders (a wide variety of selection tools could be used to make the masking very fast)

If you mean the sky, its very easy to darken it:
Do this: inspect the channels (R, G, and B - each channel contains information on how the colors are represented in your image).
As you flip through the channels, (Ctrl 1 for Red, Ctrl 2 for G, Crtl 3 for , you would notice that in the blue channel, the sky is almost completely white and the subject dark. Why is this great ? Because you need to select the sky -and if the blue channel offers the maximum contrast between your selection, its most likely you will get the best selection.

So, with the blue channel active, use the magic want, tolerance 15 (play with this no) and click on the sky. you would have selected almost all of it neatly - click a couple more time if you think you missed some. Now "feather" the selection so the edges are not coarse (color transition should be smooth) [Select->feather->3pix]

With the sky now selected, hit ctrl+~ to get back to the RGB (color) view. Ta-dah- the sky is neatly selected. Now do whatever you want (example, copy it to a new layer and set mode to multiply - looks very nice and blue now)

Again, this brings down the entire process to less than 2 mins

regds
arjun

wxwax

arjun, you're a great source, man! I'll try that right now. I did mean the sky, not the walls. Thanks for another great tutorial. I think I'm gonna copy this thread into the Hall of Wisdom.

rutt

Thanks, Sid. Thanks, Arjun. Very much. Both of you have really helped me think about this.

Arjun posted a low resolution version of his edit, so I followed his steps, perhaps imprefectly. It was a good excersize. I followed the steps from his ffirst message up until the S-Curve move. Instead of that, I converted to LAB and adjusted the background (topmost) layer to darken it and increase its saturation. Then I converted back to RGB and use shadow/highlight on the foreground (lower) level. I got this:



I think this is damn good, given the amount of work I put into it (maybe 10 minutes, 8 of which were spent learning to invert a layer mask and figuring out which kind of layer to add -- hide vs reveal).

I wanted to compare Sid's work with Arjun's method. So here are some crops from the two images for comparison.

Sid's:



Arjun's:



Sid's:



Arjun's:



I guess I could have done a better job with curves on the girl after shadow/hightlight when I followed Arjun's recipie. But that really isn't my point.

I found the hair on the girl's forhead to be the really difficult part of this image. I had problems with this both when I tried to make a mask or selection manually and when I tried global plate blending techniques. Arjun's technique works amazingly well here, give the level of effort that went into it. It isn't as good as Sid's result. Maybe a little sharpening could help? But it's much better than I was ever able to do. I suppose the the amount of blur used here directly effects the fineness of detail visible in the hair. Sid, how time consuming was it to work on this part of the image? How did you do it, exactly?

OK, now I want to be more critical. To be honest, both images look fake to me. Maybe I've just looked at this too long, but as I said, the Singapre girl in front of the harbor also looks fake and I'm sure that was a very expensive effort. Now I can tell you more about what doesn't look right.

The problem is that the background was really much brighter than the subject. The lighting and shadows around the edge of the girl show this. If I had used a flash to equalize the light, or if the girl had somehow been sunlit, the edges would have looked quite different. So when the background is artificially darkened to bring out it's detail, the lighting on the edges doesn't look right.

Look at the knees from both images. There is a bluish interior edge line. This only shows up on the edge with the background, not on the edge between the two knees. This is a shadow from the bright backlight of the original (I think). But does it look right with the new darkend background? Look at the edge between the two knees. That edge is natural and looks it.

Compare the edge between the girl's face and the background vs her nose and cheek. The highlight around the nose is natural and looks natural. The artificial edge really shows that it was silhouetted against the bright background. But the background is no longer so bright and now it shows.

I have another coment about Arjun's technique. Notice the halo it creates. You can see this better on the bottom of the front leg and around the foot. I think this comes from the blur. Is there anyway to control this? Using a smaller number for the blur makes a smaller halo, but then the edge looks even sharper and less natural.

I've learned a ton from Arjun. But I still wonder if this image can be saved. Sid came close, but it's still not what I'm looking for. Perhaps this shows that what I want just isn't possible. The defect in this image was just too extreme. Maybe the FM DRI action was right when it lightened the background.

So perhaps the conclusion is that I just should have used a flash and too bad. Or maybe there is still more fun to be had with this image? I have an idea or so to try out, but I'm still frustrated.

Sid, how much time did you spend on the mask? Be honest. I have an idea for a start up company. Suppose there were a web site where you could submit an image and some instructions (like "select the girl") and within a day you'd get back a mask that did what you wanted? How much would you pay for this service? Ok, here is my implementation: outsource it to India or China. Train people over there to do the painstaking work and pay them the going generous wage. I'm guessing one could charge a reasonable fee for this wrok and still make a nice profit.

ginger_55

What is a "mask" ? OK, forget I ever asked that. But honest to J, I do not understand jargon................. a company could be great.

Or a jargon book for 4 year olds.

g

(I signed some photos, all, last night, I did it as good as I could, the pen was losing it, well, it is of record. I am on to look up an address then have some stuff to do.....outside of the house.) (I should have taken the stuff to my daughter's, she has a great signature, and she has a name.)

wxwax

When I tried that, arjun, it seemed to give a halo around her legs. Might be the selection, could be the mask I did.

So instead, I used the Cletus Method to burn the areas I wanted darker.