Ad
Ad
Ad
Pages: [1]   Bottom of Page
Print
Author Topic: Blended exposures  (Read 3645 times)
jdemott
Sr. Member
****
Offline Offline

Posts: 434


« on: November 30, 2005, 07:29:49 PM »
ReplyReply

One of the most common situations where I want to use a blended exposure technique is when the sky is much lighter than I would like so I want to create the digital equivalent of a ND grad filter.  In the simplest case, I can accomplish that by simply doing two RAW conversions (in ACR or other RAW converter) using different exposure settings but otherwise identical conversion settings.  I then layer the two files in PS and use one of various blending techniques so that I can display the darker version of the sky and the lighter foreground.

Sometimes, particularly when there is a very sharp transition at the horizon such as a canyon wall, there will be a very small (one pixel) halo at the horizon that makes it difficult to get a nice blend.  I see this effect even when all sharpening is off and both layers were converted from the same RAW file.  The halo is clearly visible if I change the upper layer blend mode to Difference so I can see that the halo results from differences in the files and is not due to anything in my blending technique.

What causes this problem and what can I do about it?
Logged

John DeMott
Tim Gray
Sr. Member
****
Offline Offline

Posts: 2002



WWW
« Reply #1 on: December 01, 2005, 07:43:56 AM »
ReplyReply

Here's a tutorial from this site.  Sometimes I use the DRI action from Fred Miranda that referenced.  And here's another link.


The easiest way to solve the problem manually is to set up a mask on the background and then set the brush (white reveals, black conceals) to 20 or 30%opacity then paint out the layer you don't want using successive strokes.  I actually start with 100% and a very large brush  and very roughly paint out the areas approximate to the boundary (not too close) then successively make the brush smaller with less opacity as I start to paint the boundary itself.

As a refinement you can add a gradient to the mask that, although follows a straight line can be tilted depending on how you draw it to give you a rough approximation then either add to or erase (erase or white) the parts of the gradient that aren't working.

Or, if you know you've got a problem in the field, and the histogram should say so, you can bracked and use HDR - problem with movement, obviously.
Logged
bobrobert
Guest
« Reply #2 on: December 01, 2005, 08:03:20 AM »
ReplyReply

Quote
Here's a tutorial from this site.  Sometimes I use the DRI action from Fred Miranda that referenced.  And here's another link.
The easiest way to solve the problem manually is to set up a mask on the background and then set the brush (white reveals, black conceals) to 20 or 30%opacity then paint out the layer you don't want using successive strokes.  I actually start with 100% and a very large brush  and very roughly paint out the areas approximate to the boundary (not too close) then successively make the brush smaller with less opacity as I start to paint the boundary itself.

As a refinement you can add a gradient to the mask that, although follows a straight line can be tilted depending on how you draw it to give you a rough approximation then either add to or erase (erase or white) the parts of the gradient that aren't working.

Or, if you know you've got a problem in the field, and the histogram should say so, you can bracked and use HDR - problem with movement, obviously.
[a href=\"index.php?act=findpost&pid=52552\"][{POST_SNAPBACK}][/a]
Logged
jdemott
Sr. Member
****
Offline Offline

Posts: 434


« Reply #3 on: December 01, 2005, 12:59:41 PM »
ReplyReply

Tim,

Thanks for replying.  I probably didn't explain my question very clearly.  I am familiar with those blending tutorials (and others).  Below is a sample of what I am talking about, with a 100 percent crop taken from an unsharpened D2X file converted in ACR.  The first two frames show the same file converted with identical settings other than a 1-1/2 stop difference in exposure.  The final frame shows the darker frame superimposed over the lighter one in Difference mode, illustrating that the images are not congruent along the horizon, even though they are the same image.  The sky is one pixel larger in width in the dark version than the light one.

When I blend the two layers, it is usually possible to more or less obscure the "halo" effect either by using the gaussian blur effect or a soft edged brush or a gradient.  Of course, the ability to obscure the halo comes from muddying the waters and using some of each image (or not blending at all near the areas of transition).  The more one wants to take full advantage of the exposure differences between the two images (i.e., to present a properly exposed foreground right up to the horizon and a sky that is not overexposed right down to the horizon) the more difficult it is to obscure the halo.  It can make for some fairly laborious editing of the mask in some situations.

In any event, assume I successfully obscure the halo by using some sort of blurred or soft-edged layer mask for blending.  There is still some discontinuity in tonality from the halo present in the blended version but it isn't readily apparent because it it lurking just below our visual threshold.  Unfortunately, that lurking halo can re-appear at the final stages of processing when it is time to sharpen the image.  Sharpening of course accentuates any differences in tonality along edges in the image.  The result often is an unrealisitic, oversharpened appearance along the horizon when I apply sharpening settings that are appropriate for the overall image.

I do often use the technique of making multiple exposures in the field.  Of course the greater the exposure difference between the two images, the more apparent any halo will be.  I deliberately chose the example of multiple conversion from one RAW file because I knew people would assume I had a problem with camera movement between multiple exposures.  (I've experimented with HDR quite a bit, but I don't think the alignment feature works very well and I have yet to get results from an HDR that are as good overall as one of the more conventional exposure blending techniques.  That's a topic for another discussion however.)

Anyway, I remain puzzled about why two conversions of the same file produce incongruent results and I am open to suggestions for easy ways to deal with it.

[attachment=43:attachment]
Logged

John DeMott
bobrobert
Guest
« Reply #4 on: December 02, 2005, 04:37:32 AM »
ReplyReply

Quote
Tim,

Thanks for replying.  I probably didn't explain my question very clearly.  I am familiar with those blending tutorials (and others).  Below is a sample of what I am talking about, with a 100 percent crop taken from an unsharpened D2X file converted in ACR.  The first two frames show the same file converted with identical settings other than a 1-1/2 stop difference in exposure.  The final frame shows the darker frame superimposed over the lighter one in Difference mode, illustrating that the images are not congruent along the horizon, even though they are the same image.  The sky is one pixel larger in width in the dark version than the light one.

When I blend the two layers, it is usually possible to more or less obscure the "halo" effect either by using the gaussian blur effect or a soft edged brush or a gradient.  Of course, the ability to obscure the halo comes from muddying the waters and using some of each image (or not blending at all near the areas of transition).  The more one wants to take full advantage of the exposure differences between the two images (i.e., to present a properly exposed foreground right up to the horizon and a sky that is not overexposed right down to the horizon) the more difficult it is to obscure the halo.  It can make for some fairly laborious editing of the mask in some situations.

In any event, assume I successfully obscure the halo by using some sort of blurred or soft-edged layer mask for blending.  There is still some discontinuity in tonality from the halo present in the blended version but it isn't readily apparent because it it lurking just below our visual threshold.  Unfortunately, that lurking halo can re-appear at the final stages of processing when it is time to sharpen the image.  Sharpening of course accentuates any differences in tonality along edges in the image.  The result often is an unrealisitic, oversharpened appearance along the horizon when I apply sharpening settings that are appropriate for the overall image.

I do often use the technique of making multiple exposures in the field.  Of course the greater the exposure difference between the two images, the more apparent any halo will be.  I deliberately chose the example of multiple conversion from one RAW file because I knew people would assume I had a problem with camera movement between multiple exposures.  (I've experimented with HDR quite a bit, but I don't think the alignment feature works very well and I have yet to get results from an HDR that are as good overall as one of the more conventional exposure blending techniques.  That's a topic for another discussion however.)

Anyway, I remain puzzled about why two conversions of the same file produce incongruent results and I am open to suggestions for easy ways to deal with it.

[attachment=43:attachment]
[a href=\"index.php?act=findpost&pid=52599\"][{POST_SNAPBACK}][/a]

If the foreground is a mask cant you just expand the mask by 1 pixel ( or more )?
PS> modify> expand or contract depending on what is selected???
Logged
Tim Gray
Sr. Member
****
Offline Offline

Posts: 2002



WWW
« Reply #5 on: December 02, 2005, 07:55:47 AM »
ReplyReply

The boundary between the dark rock and bright sky isn't a 1 dimensional line, it has width due to the antialiasing fileter, and lens blur (but obviously not camera shake or registration since it's the same shot).  As you create a over and under layer and then mask and blend (without blur or perhaps feather) you won't be able to avoid that boundary region because it was there in the beginning.  Just magnify at say 800% and you'll see the true nature of the boundary.
 
I have struggled for hours in the past dealing with this on a pixel by pixel basis, but the techniques you are already familiar with basically solve the problem.  I have found that even bluring the selection isn't usually a satisfactory solution since it usually gives rise to sever halos.  Using the painting and gradient techniques I've never encountered an objectionable artefact in a print after final sharpening (PK sharpener).
« Last Edit: December 02, 2005, 07:57:19 AM by Tim Gray » Logged
jdemott
Sr. Member
****
Offline Offline

Posts: 434


« Reply #6 on: December 02, 2005, 12:29:45 PM »
ReplyReply

Quote
The boundary between the dark rock and bright sky isn't a 1 dimensional line, it has width due to the antialiasing fileter, and lens blur (but obviously not camera shake or registration since it's the same shot). As you create a over and under layer and then mask and blend (without blur or perhaps feather) you won't be able to avoid that boundary region because it was there in the beginning. Just magnify at say 800% and you'll see the true nature of the boundary.


Of course, because of anti-aliasing, the boundary between light and dark will have an intermediate zone of gray (or medium tone).  What is interesting to me is that when you change the exposure setting in the RAW conversion, then the transition zone is, in effect, displaced spatially.  To over-simplify the situation, when you adjust the exposure setting downward, you would expect all pixels to be darkened more or less proportionately--those that are in the bright sky would be darkened to about the same extent as would the dark foregound, which would also be about the same extent as the intermediate zone.  If that were the case, then the Difference mode blend of the two images would reveal essentially a uniform gray (or middle tone). But, as the Difference mode portion of my example shows, the exposure adjustment is not uniform across the boundary between light and dark, resulting in  "halo."  I assume that the effect results from some interplay of how the de-mosaicing algorithms handle color information (since brightness and color are intertwined in an RGB system) and the fact that exposure adjustments aren't uniform throughout the range of values (when we move the exposure slider upward, the effect we usually want to achieve is to increase the brightness of most of the image without necessarily losing the highlights altogether).

Quote
Using the painting and gradient techniques I've never encountered an objectionable artefact in a print after final sharpening


Usually, the gradient technique is my preferred method--it is simple to apply and it produces results that people are comfortable with visually because it so closely mimics a ND grad filter.  Of course it doesn't work as well when the boundary between light and dark isn't straight (such as is seen on a small scale in the example I posted).  Then I generally use the painting technique.  Where I struggle the most is images in which the sky and the foreground both need proper exposure right up to the boundary--for example a rocky outcropping caught in morning sunlight against a dramatic sky.  Sigh, I guess that I'll just have to wait until Nikon makes the D5X with an 11 stop dynamic range.  
Logged

John DeMott
Peter McLennan
Sr. Member
****
Offline Offline

Posts: 1695


« Reply #7 on: December 02, 2005, 01:04:50 PM »
ReplyReply

Quote
Of course, because of anti-aliasing, the boundary between light and dark will have an intermediate zone of gray (or medium tone).

Camera anti-aliasing?  Or in ACR?  

If it's an ACR issue, maybe Adobe can help us out in a later rev.  

I've experienced the same thing with dark, bare trees against a hot sky.  No way I'll ever be able to paint out the boundary, it's just too complex.

Great discussion

Peter
Logged
jdemott
Sr. Member
****
Offline Offline

Posts: 434


« Reply #8 on: December 02, 2005, 03:32:08 PM »
ReplyReply

Quote
Camera anti-aliasing? Or in ACR?

Well, the camera will always have anti-aliasing, unless you have something like one of the Kodak DSLRs that don't have an anti-aliasing filter. I don't know whether the ACR effect that I've described would be called an anti-aliasing effect or more likely a de-mosaicing or exposure adjustment effect.  
Quote
I've experienced the same thing with dark, bare trees against a hot sky. No way I'll ever be able to paint out the boundary, it's just too complex.

Your best bet for that type of situation is a contrast mask solution, such as was described in the link that was mentioned earlier in this thread
Logged

John DeMott
Tim Gray
Sr. Member
****
Offline Offline

Posts: 2002



WWW
« Reply #9 on: December 02, 2005, 03:40:05 PM »
ReplyReply

Quote
What is interesting to me is that when you change the exposure setting in the RAW conversion, then the transition zone is, in effect, displaced spatially. ....you would expect all pixels to be darkened more or less proportionately, those that are in the bright sky would be darkened to about the same extent as would the dark foregound, which would also be about the same extent as the intermediate zone. [a href=\"index.php?act=findpost&pid=52683\"][{POST_SNAPBACK}][/a]

so you're suggest that the layers appear out of registration?

also,  "darkened to about the same extent" - well, I agree the changes will be proportional, but in absolute terms, the boundary will not be changed by the same amount as the ground or sky.  I'll look when I get home, but I don't think  histogram just shifts unchanged left or right as you change exposure, the overall shape changes somewhat as well.

you could easily take the raw converter out of the equation - just convert to tiff and create an over ex layer and an under ex layer and take a look
« Last Edit: December 02, 2005, 03:45:27 PM by Tim Gray » Logged
jdemott
Sr. Member
****
Offline Offline

Posts: 434


« Reply #10 on: December 02, 2005, 06:19:58 PM »
ReplyReply

Quote
so you're suggest that the layers appear out of registration?

They appear somewhat similar to what you see when layers are out of registration.  If you duplicate a layer and change the blend mode of the upper layer to Difference, then the screen will be black.  If you displace the upper layer by one or two pixels (using the Move tool) while still in Difference mode, you will see a halo similar to what I show in my example.  I often use the Difference mode to check registration.

Quote
you could easily take the raw converter out of the equation - just convert to tiff and create an over ex layer and an under ex layer and take a look

Good idea.  I just tried that and the results were interesting but not conclusive.  I used the same file that I used for the posted example and created three layers.  The Background layer is a straight RAW conversion from ACR--the foreground is normally exposed and the sky is almost but not quite blown out.  Layer 1 is a second RAW conversion with the exposure adjusted down 2 stops--the foreground has become darker and ACR has done a nice job recovering highlights in the sky so there are interesting looking clouds and some blue sky (not as nice as a properly exposed sky but okay).  Layer 2 is a duplicate of the Background layer adjusted in PS to reduce the exposure by 2 stops (Image>Adjust>Exposure)--the foreground nearly matches Layer 1, but the sky is simply a darker version of the washed out sky in the background layer, an ugly gray mess.  Comparing the results in Difference mode, I see the halo when I compare Layer 1 and the Background but not when I compare Layer 2 and the Background.  The reason I say that the results are not conclusive is that the sky area in Layer 2 remains fairly featureless and in Difference mode there is a clear border between sky and foreground but not a halo--it is possible (but I don't think it likely) that there is some small displacement but you don't see the halo because the sky is so featureless.
Logged

John DeMott
Pages: [1]   Top of Page
Print
Jump to:  

Ad
Ad
Ad