Noise gathering?

Discussion forum for Tawbaware's TuFuse and TuFuse Pro software
dsjtecserv
Posts: 551
Joined: Sat May 05, 2007 3:25 pm
Location: Northern Virginia
Contact:

Noise gathering?

Post by dsjtecserv » Sun Apr 20, 2008 11:50 am

I had some expectation that focus stacking would have a side benefit of reducing noise, as is the case for other stacking techniques. But it seems that TuFuse (and presumably any focus-stacking software) is actually quite effective at gathering the best, sharpest noise from each of the tributary images! That makes sense, because what TuFuse does for a living is find the sharpest pixels with the most contrast from their neighbors; unfortunately that would include noise.

This hasn't been an observable problem in my previous TuFuse work, but it became very evident when I accidentally set my ISO to 1600 and tried a 16-image stack. Even though I used the noise reduction in ACR to reduce the noise in each of the images to a quite acceptable level, TuFuse found and included the "best noise" from each of the images in the final output. The 100% crops below show one of the input images and the same area from the TuFuse output.

Image

I don't see any setting in TuFuse that could be used to combat this, but does anyone have any idea for how the source images could be massaged to minimize this? (other than paying attention to the ISO in the first place, and using noise reduction, which was already used.) (Edit: I should also note that the noise in the TuFuse image is so granular that the CS3 noise reduction tool, even at its highest setting, doesn't make a significant dent in the noise. It evidently views the noise as part of the picture.)

Dave

maxlyons
Posts: 3649
Joined: Fri Jun 20, 2003 8:55 pm
Location: USA
Contact:

Re: Noise gathering?

Post by maxlyons » Sun Apr 20, 2008 2:19 pm

dsjtecserv wrote:I had some expectation that focus stacking would have a side benefit of reducing noise, as is the case for other stacking techniques. But it seems that TuFuse (and presumably any focus-stacking software) is actually quite effective at gathering the best, sharpest noise from each of the tributary images! That makes sense, because what TuFuse does for a living is find the sharpest pixels with the most contrast from their neighbors; unfortunately that would include noise.
Dave,

You've summarized the problem nicely..."noise" is the same as "detail" from TuFuse's perspective. In fact, I've been working for the last week or so on incorporating noise reduction capabilities into TuFuse, trying to distinguish between "noise detail" and "detail detail". I've made some progress, and hope to have something as part of the next release, but no noise reduction algorithms are perfect. As you reduce noise you end up reducing "real" detail as well.

The best advice I can give it to start with the best images possible...ISO 1600 probably won't give the best results.

Max

dsjtecserv
Posts: 551
Joined: Sat May 05, 2007 3:25 pm
Location: Northern Virginia
Contact:

Post by dsjtecserv » Sun Apr 20, 2008 2:56 pm

Thanks, Max. I was afraid that would be your answer, but I'm not surprised. Glad to hear that you are working on a NR algorithm. If it can distinguish between detail I want and detail I don't want that will indeed be remarkable!

Dave

maxlyons
Posts: 3649
Joined: Fri Jun 20, 2003 8:55 pm
Location: USA
Contact:

Post by maxlyons » Sun Apr 20, 2008 3:19 pm

dsjtecserv wrote:Thanks, Max. I was afraid that would be your answer, but I'm not surprised. Glad to hear that you are working on a NR algorithm. If it can distinguish between detail I want and detail I don't want that will indeed be remarkable!
Yes, that would be remarkable!

I think it will help, but doubt it will solve all noise problems. I think the old adage will still apply: "there ain't no such thing as a free lunch".

Max

dsjtecserv
Posts: 551
Joined: Sat May 05, 2007 3:25 pm
Location: Northern Virginia
Contact:

Post by dsjtecserv » Sun Apr 20, 2008 3:36 pm

No, I was engaging in delusional thinking, of course. Even if TuFuse can incrementally reduce the amount of noise added at the time of fusion that would be helpful. As I said, I had not noticed additional noise (although there might in fact have been some) in TuFuse output of stacks shot at ISO 100 and 200.

Dave

Karsten33
Posts: 9
Joined: Sat Mar 22, 2008 9:33 am

Post by Karsten33 » Mon Apr 21, 2008 9:17 am

Thanks Max,

I have read this with interest:
You've summarized the problem nicely..."noise" is the same as "detail" from TuFuse's perspective. In fact, I've been working for the last week or so on incorporating noise reduction capabilities into TuFuse, trying to distinguish between "noise detail" and "detail detail". I've made some progress, and hope to have something as part of the next release, but no noise reduction algorithms are perfect. As you reduce noise you end up reducing "real" detail as well.
In last time I have tried Photoacute Demo, which now have also focus stacking support and it seams, that they have solved the noise removal problem ( and resolution enhancement ) quite well.

I wish your success for your noise removing work on TuFuse and I wait impatiently :D

regards Karsten

rickb
Posts: 140
Joined: Fri Jul 11, 2003 9:20 pm
Location: SF Bay area
Contact:

Post by rickb » Mon Apr 21, 2008 10:06 am

I noticed the same thing, but Tufuse is not alone in gathering noise.
Since, Tufuse can do both exposure HDR and focus blending, is there a way to tell it to avg xyz frames first and then do focus stacking? Then you can reduce the noise first, and second, look for the best focus. If possible, you could take multiple ISO1600 exposures at each focus, then focus stack the reduced noise averaged images.
In CS3, HDR does reduce the noise, but of course, it does not do focus blending. With Tufuse, there is the possibility of doing more.

thanks for the great new SW tools...

trpta
Posts: 37
Joined: Mon Jul 14, 2003 3:11 pm

Post by trpta » Mon Apr 21, 2008 6:26 pm

rickb wrote:Since, Tufuse can do both exposure HDR and focus blending, is there a way to tell it to avg xyz frames first and then do focus stacking? Then you can reduce the noise first, and second, look for the best focus.
In principle, the command line options for Tufuse already allow for this. For example, in theory ...

Code: Select all

tufuse -p 2 --wMode1 0 --wContrast1 0 --wExposure1 1 --wSaturation1 0 --wMode2 1 --wContrast2 1 --wExposure2 0 --wSaturation2 0 *.tif
... should do weighted-average exposure blending on the first pass and max contrast stacking on the second pass. I haven't tested this yet, but I suspect that it won't quite work as intended right now. Tufuse needs to have a way to do exposure blending in the first pass only on images with the same or similar focus distance. I don't know whether Tufuse currently contains that logic. If not, maybe Max could consider adding it by referring to the focus distance information in the EXIF headers of the source files.

If Tufuse can't do an automatic two-stage pass with exposure blending before focus stacking, then you could accomplish the same thing manually by first running this command on each series of images that have a common focus distance:

Code: Select all

tufuse -p 1 --wMode 0 --wContrast 0 --wExposure 1 --wSaturation 0 *.tif
... then perform focus stacking on the resulting exposure-blended files with this:

Code: Select all

tufuse -p 1 --wMode 1 --wContrast 1 --wExposure 0 --wSaturation 0 *.tif
Max, can Tufuse handle this two-stage process automatically?

Taylor

maxlyons
Posts: 3649
Joined: Fri Jun 20, 2003 8:55 pm
Location: USA
Contact:

Post by maxlyons » Mon Apr 21, 2008 7:18 pm

trpta wrote:Max, can Tufuse handle this two-stage process automatically?
No. In order to do so it would have to be able to determine which shots were focused at the same distance. This is not simple...in fact, I doubt it is even possible to do with 100% reliability by simply examining the image.

Looking at EXIF data seems attractive, but it is (a) not always present in the image, (b) frequently doesn't contain the subject distance field and (c) frequently incorrect, so TuFuse doesn't use it for any purpose.

I think that the steps you've outlined will work. One could always automate the steps by creating a simple batch file to invoke TuFuse in sequence with the appropriate images.

Max

trpta
Posts: 37
Joined: Mon Jul 14, 2003 3:11 pm

Post by trpta » Wed Apr 23, 2008 6:30 pm

Max,

PTAssembler allows you to manually specify stacking groups for your source images if you don't want to rely on the automatic selection mechanism. Could you do something similar in Tufuse Pro that would faciliate the "manual" two-stage process described in my earlier posting? Allow the user to manually specify "focus" groups of images with the same or similar focus distance. Then Tufuse Pro could exposure blend these groups in the first pass, followed by focus stacking the results in the second pass. What do you think?

Taylor

maxlyons
Posts: 3649
Joined: Fri Jun 20, 2003 8:55 pm
Location: USA
Contact:

Post by maxlyons » Wed Apr 23, 2008 10:23 pm

trpta wrote:Could you do something similar in Tufuse Pro that would faciliate the "manual" two-stage process described in my earlier posting?
I'll consider it. However, I'm reluctant to add more complexity than necessary to the GUI. It isn't clear to me why one would take a series of images at such a high ISO that requires noise reduction (ISO 1600) for focus blending. I can't imagine a lot of people taking a series of images for focus blending purposes at ISO 1600 (which would then require capturing a large number of additional images for noise reduction purposes before focus blending).

Max

trpta
Posts: 37
Joined: Mon Jul 14, 2003 3:11 pm

Post by trpta » Wed Apr 23, 2008 11:26 pm

Actually, I wasn't thinking of this as something I would only do to deal with high ISO images. I was considering that as a general matter I might want to do the exposure blending before the focus stacking for all of my images. That way, the image noise would be reduced and the focus stacking would always operate on well exposed images. It's not critical that TuFuse Pro be able to do this, of course, since I can always write a batch file to accomplish the same result using TuFuse. Just thought it might be a nice feature. Thanks for listening.

Taylor

dsjtecserv
Posts: 551
Joined: Sat May 05, 2007 3:25 pm
Location: Northern Virginia
Contact:

Post by dsjtecserv » Thu Apr 24, 2008 8:15 am

maxlyons wrote: I'll consider it. However, I'm reluctant to add more complexity than necessary to the GUI. It isn't clear to me why one would take a series of images at such a high ISO that requires noise reduction (ISO 1600) for focus blending. I can't imagine a lot of people taking a series of images for focus blending purposes at ISO 1600 (which would then require capturing a large number of additional images for noise reduction purposes before focus blending).
Well, I can think of at least one person who has done this!

While it wasn't my intent to shoot at 1600, it would be nice to be able to rescue the series by reducing the noise. And, truth be told, I'm not sure there aren't situations where this might be an intentional decision. In the case at hand, the light was fading and even the small breeze resulted in subject motion. I'm not sure I could have gotten useful images at a lower ISO. And the noise at 1600 isn't bad with routine noise reduction. If the fused image had the noise of the original images, I'd be satisfied. But the problem is that the software collects the best noise from all the images, which makes the summ of the parts unacceptable.

I'm not saying the solution offered by trpta is the best approach, but some method of reducing the noise gathering effect would be welcome. It is unlikely that I would have been able or interested in shooting several more layers for each shot to allow for exposure blending. The method you mentioned earlier, to attempt to distinguish between noise and image detail in the focus-blended images themselves, would be needed for most of these situations , although it sounds mighty hard to do.

Dave

maxlyons
Posts: 3649
Joined: Fri Jun 20, 2003 8:55 pm
Location: USA
Contact:

Post by maxlyons » Thu Apr 24, 2008 8:26 am

dsjtecserv wrote:Well, I can think of at least one person who has done this!
I understand...but I doubt that this will be a widely enough adopted strategy that it merits adding the additional complexity into the GUI to handle this special case. In any case, one can simply run the GUI twice...a second time for the results of the exposure blending first pass.
I'm not sure there aren't situations where this might be an intentional decision. In the case at hand, the light was fading and even the small breeze resulted in subject motion. I'm not sure I could have gotten useful images at a lower ISO.
I see the allure of doing this, but if the breeze was sufficient to cause subject motion, then won't you end up with misalignments during the exposure blending stage? If the breeze was such that an exposure at ISO 100 was not feasible, wouldn't it also preclude taking 2 (or more) exposures at higher ISOs and expecting them to line up correctly? Or, was it the case the breeze would stop for long enough to allow a X second exposure, but not a 2*X exposure, and during the moments when the breeze stopped, the subject returned to *exactly* the same position every time so that exposure blending was feasible?

Max

rickb
Posts: 140
Joined: Fri Jul 11, 2003 9:20 pm
Location: SF Bay area
Contact:

Post by rickb » Thu Apr 24, 2008 10:02 am

Another application for me is combining IR photos taken at ISO100 with a tripod...
Often there is a certain amount of noise to remove and I can do that first with, for example, CS3 HDR and save the file. Then I can combine these saved noise reduced images with stitching and blending. It would be nice to have a more automatic way of doing this process, but I realize, there might not be much demand yet for noise reduction first and focus blending or stitching second, given most people can shoot low noise images to start with.

Another case is taking color photos at 1:1 or greater, with a good DSLR mounted on a tripod and using natural light. That situation often requires either a faster shutter speed, like 1/100 sec, or a very slow one, 1 sec and slower. The reason is the vibration of the shutter, you can do mirror lockup first, but when the first shutter curtain fires, there is vibration, and in macro that means blurring. So, sometimes I use high ISO settings to get sharper images, then I use noise reduction with HDR, or averaging, save the resuluts and do focus blending on those.

thanks for all the fun imaging tools.

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest