Deprecated: preg_replace(): The /e modifier is deprecated, use preg_replace_callback instead in /www/htdocs/w0068b6d/smf/Sources/Load.php(225) : runtime-created function on line 3
local contrast adjustment independent from compression ratio
forum@fdrtools.com
June 20, 2019, 08:59:24 AM *
Welcome, Guest. Please login or register.

Login with username, password and session length
News: SMF - Just Installed!
 
   Home   Help Search Login Register  
Pages: [1]
  Print  
Author Topic: local contrast adjustment independent from compression ratio  (Read 19571 times)
Carel
Resident
***
Posts: 51


View Profile
« on: July 09, 2006, 07:33:02 PM »

I dont know how much the LC is tied to the compression ratio with your TMO, but it would be nice to be able to select them separately. Sometimes I would like to set the compression a bit higher, but the local contrast seems to get too strong. At other times one would like to try higher LC without increasing the compression. Ofcourse I can always add LC in photoshop, but that usually destroys the detail in the highlights. Applying LC with the TMO does not do this. (Does this make sense...?)
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #1 on: July 10, 2006, 09:18:04 AM »

Quote
I dont know how much the LC is tied to the compression ratio with your TMO, but it would be nice to be able to select them separately. Sometimes I would like to set the compression a bit higher, but the local contrast seems to get too strong. At other times one would like to try higher LC without increasing the compression. Ofcourse I can always add LC in photoshop, but that usually destroys the detail in the highlights. Applying LC with the TMO does not do this. (Does this make sense...?)

I think I know what you mean. Compression and contrast are indeed bound together: increasing compression increases contrast and vice versa. The problem is that this currently happens on a global scale and the effect of changing compression on a certain local spot can be surprising and unwanted. In other words: there is not enough control on how compression should be applied to different regions of an image. My intention is to tackle this problem with "selective tone mapping" which means that you shall be able to determine a certain image region and specify individual tone mapping settings for this region. Unfortunately this is nothing that will come tomorrow.
Logged
Carel
Resident
***
Posts: 51


View Profile
« Reply #2 on: July 10, 2006, 09:33:42 PM »

Quote
My intention is to tackle this problem with "selective tone mapping" which means that you shall be able to determine a certain image region and specify individual tone mapping settings for this region.


Yes, that is what I call the "windows" problem and it is a major one for all TMOs.

But in my suggestion above, I was just talking about the global decoupling of LC and compression.
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #3 on: July 10, 2006, 09:59:18 PM »

Quote
Yes, that is what I call the "windows" problem and it is a major one for all TMOs.

But in my suggestion above, I was just talking about the global decoupling of LC and compression.

Hmmmh, maybe this is a bit abstract. I think a practical example could make it more clear to me...
Logged
Carel
Resident
***
Posts: 51


View Profile
« Reply #4 on: July 11, 2006, 03:19:25 AM »

What I mean is that sometimes one wants to compress the highlights and the shadows to a certain level, but with less, or in other instances more, local contrast than what FDRTools provides. At the moment one would do this by choosing a lower compression value and changing the contrast up or down. But from working with HDR compression schemes that do not provide local contrast enhancement I know that sometimes one gets better results with low LC. For instance in a room with windows, one usually needs a high compression to keep detail in the brightest areas, but at the same time one wants to avoid the "dirty walls" effect that high local contrast brings out.
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #5 on: July 11, 2006, 09:05:52 AM »

Quote
What I mean is that sometimes one wants to compress the highlights and the shadows to a certain level, but with less, or in other instances more, local contrast than what FDRTools provides. At the moment one would do this by choosing a lower compression value and changing the contrast up or down. But from working with HDR compression schemes that do not provide local contrast enhancement I know that sometimes one gets better results with low LC. For instance in a room with windows, one usually needs a high compression to keep detail in the brightest areas, but at the same time one wants to avoid the "dirty walls" effect that high local contrast brings out.

I think we are talking about the same thing namely how to tell an algorithm what to do. In this case you would say: "I want more details in the window area but don't accentuate the interior details too much." In FDRTools you can currently move the two sliders "Compression" and "Contrast" to tell what you want. This is easy enough and gives nice results for many scene types but is not adequate for some others like the one you mentioned.

Now what to do? If I get you right your suggestion is to have two independently acting sliders: one for overall compression and one for local contrast. The problem with this is: I currently have no idea how this could reliably function in all thinkable situations Sad

Hence my answer to this problem will be "selective tone mapping". It will allow to express your wishes in an easy an understandable way and I believe it will be general enough to cope with all thinkable situations.
Logged
Carel
Resident
***
Posts: 51


View Profile
« Reply #6 on: July 11, 2006, 07:05:09 PM »

I am very curious how you plan to implement selective tonemapping. To get back to those blasted windows: The problem in such a case is that there is always a very sudden change. So If one were to select the area you are right back to masking the window panes, plants in front of the window, sheer (see-though) curtains and all those other nightmares of masking (at least for me).

My favorite method at the moment is to not mask out the window beforehand, but to make a separate image optimized for interior and one for exterior, layer the exterior over the interior, apply mask>hide all and then rub through just the view through the windows. It works better for me this way than masking beforehand, because one can see right away what areas at the very edge of the window-panes need to be included or excluded and correct on the spot. I even use a brush in the shape of a squeegee for this, so it is like washing dirty windows Cheesy
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #7 on: July 11, 2006, 08:49:34 PM »

Quote
I am very curious how you plan to implement selective tonemapping.

So am I Cheesy No, it will be much like working with masks in photoshop. Select a region with a brush tool, select tone mapping parameters for the region and see how it looks like. Apply the same procedure to other regions then if necessary.

Quote
To get back to those blasted windows: The problem in such a case is that there is always a very sudden change. So If one were to select the area you are right back to masking the window panes, plants in front of the window, sheer (see-though) curtains and all those other nightmares of masking (at least for me).

There will definitely no problems with eye-catching seams or gradients between regions with differing tone mapping settings.

Quote
My favorite method at the moment is to not mask out the window beforehand, but to make a separate image optimized for interior and one for exterior, layer the exterior over the interior, apply mask>hide all and then rub through just the view through the windows. It works better for me this way than masking beforehand, because one can see right away what areas at the very edge of the window-panes need to be included or excluded and correct on the spot. I even use a brush in the shape of a squeegee for this, so it is like washing dirty windows Cheesy

That sounds good. I think I would do it in a similiar way.
Logged
Carel
Resident
***
Posts: 51


View Profile
« Reply #8 on: July 14, 2006, 06:05:51 PM »

I have been thinking some more about the TMO problem of reduced contrast in the highlights after compression. Rather than implementing some kind of masking scheme, should it not be possible to deal with this during HDR assembly, by treating the darker images (for the highlights) differently from the lighter ones? Maybe the contrast of the darker images needs to be increased, or their levels shifted in some other way.

At first I thought the reduced contrast had something to do with the local contrast, which seems to "look" at the image globally. But compression schemes that dont use local contrast have the same problem. So it must have something to do with the levels of the different exposures during hdr assembly. And because it is a very "predicatable" effect, it might be possible to compensate for...?
Logged
Carel
Resident
***
Posts: 51


View Profile
« Reply #9 on: July 14, 2006, 06:56:03 PM »

To partially aswer my own question:
Yes, by manipulating the levels of the darker images (sliding up black level on darkest, sliding up black level and slightly reducing white level on next dark image and doing same on next darkest from that) I get a much better "view through the windows" This was done in photoshop, *before* assembling the hdr in FDRTools. It also produces less light halo at the edges of the window panes.  I think what this does is to "force" FDRTools to choose a different part of the image's histograms. At the moment this could not be accomplished within FDRTools, because there one can only manipulate the right "side" of each image's histogram.

None of this means that I know what I am doing Cheesy

Carel
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #10 on: July 14, 2006, 09:15:57 PM »

Quote
To partially aswer my own question:
Yes, by manipulating the levels of the darker images (sliding up black level on darkest, sliding up black level and slightly reducing white level on next dark image and doing same on next darkest from that) I get a much better "view through the windows" This was done in photoshop, *before* assembling the hdr in FDRTools. It also produces less light halo at the edges of the window panes.  I think what this does is to "force" FDRTools to choose a different part of the image's histograms. At the moment this could not be accomplished within FDRTools, because there one can only manipulate the right "side" of each image's histogram.

Hmmmh, what you achieve by moving the blackpoint to the right is effectively a contrast boost. You cut off the lower part of the spectrum and the remaining part of the spectrum is "dispersed" yielding increased overall contrast. Doing this for several layers and selecting an intensity band for each layer in FDRTools lets you assemble an HDR image with differing contrast characteristics among the intensity bands. This is a good idea! And more over it can be implemented without much efforts...

If you like I can assemble a "special edition" for you to play around a bit. Each layer would have an additional slider for blackpoint adjustment. What do you think?

Developping this idea a bit further leads to the concept of intensity band based tone mapping, i.e. providing a set of tone mapping parameters for each layer. This should not be too hard too implement either and is definitely worth a try...
Logged
Carel
Resident
***
Posts: 51


View Profile
« Reply #11 on: July 15, 2006, 02:21:32 AM »

Yes, I would love to try such a "special edition".

I got the idea for trying this when someone mentioned that digital cameras rarely produce an absolute black. I took this to mean that each conversion process would therefore interpret black differently depending on the overall exposure of the image.  So by narrowing the histogram space alocated to black, one would force the program to select the proper areas. I have no idea if any of this makes any sense, but the result of the test was convincing.

If any of this could work in a reasonably predictable fashion one could avoid having to implement a masking scheme.
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #12 on: July 15, 2006, 08:07:54 AM »

Quote
Yes, I would love to try such a "special edition".

Fine. I'm going to take up on it as soon as v1.7 is out.

Quote
I got the idea for trying this when someone mentioned that digital cameras rarely produce an absolute black.

They never do. There is always a noise signal there produced by the electronic circuits. The visual result of this noise are the ugly red, green and blue speckles in the dark areas of an image. This noise limits the usable dynamic range of a digital camera.

Quote
I took this to mean that each conversion process would therefore interpret black differently depending on the overall exposure of the image.  So by narrowing the histogram space alocated to black, one would force the program to select the proper areas.

I can only speek for FDRTools. FDRTools by intention ignores the noisy region of an image and replaces it with the data of a longer exposed image. Because of that it doesn't do any damage if we cut off the left end of the histogram. However, such a blackpoint shift in effect is a kind of tone mapping, hence means that the HDR image will be composed from several pieces of already tone mapped image data.
Logged
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.11 | SMF © 2006-2009, Simple Machines LLC Valid XHTML 1.0! Valid CSS!