DPChallenge: A Digital Photography Contest You are not logged in. (log in or register) 
 

DPChallenge Forums >> Photography Discussion >> A new (for me) noise reducer- Median blending
Pages:  
Showing posts 1 - 25 of 27, (reverse)
AuthorThread
 05/29/2013 01:37:24 PM · #1
To quote from the article linked-
"In image averaging, I had previously blended images together by averaging the value of each pixel in an image to produce something entirely new. The way that this process relates to astrophotography is that the general method is commonly used as a means of noise reduction. It becomes interesting when you realize that you donĂ¢€™t have to be blending images or taking photos of the cosmos to benefit from the same methods of noise reduction. It works incredibly well for any images that are relatively static."
Link
 05/29/2013 02:07:51 PM · #2
i've seen this technique used before (for removing non-static objects), but the caveat for noise removal is that its often isn't a viable alternative since most people who shoot at high ISO dont have static scenes.

05/29/2013 02:30:07 PM · #3
Originally posted by BrennanOB:
It works incredibly well for any images that are relatively static."

I was hoping this would help with images shot for eventual HDR merging, but from reading the description it looks as though the "extreme" exposures would simply be discarded as "spurious" data ... :-(

Originally posted by Mike:
i've seen this technique used before (for removing non-static objects), but the caveat for noise removal is that its often isn't a viable alternative since most people who shoot at high ISO dont have static scenes.

Some of my better still-lifes are shot with realtively weak lights and long exposures, which also introduces some noise, although not as much as raising the ISO. This might help with that type of shot. These were all shot with exposures of either 1 or 1.3 seconds.
05/29/2013 02:34:06 PM · #4
Hmmm ... I guess this might work for HDR if one could shoot (for example) three frames at each exposure level, do a median blend of each of those sets, and then do an HDR merge of the resultant (hopefully) cleaner images.
 05/29/2013 02:45:47 PM · #5
i dont think your camera does it but with hdr and long exposures, dslrs usually have long exposure noise reduction ability.
 05/29/2013 02:53:36 PM · #6
Funny that you bring up median blending because I just saw an article talking about it yesterday but in a completely different context. It was for removing tourists from scenes by taking maybe 15 shots over time (on a tripod) and then using median blending to remove the objects that are different (ie. the people).
 05/29/2013 03:01:30 PM · #7
Originally posted by DrAchoo:
Funny that you bring up median blending because I just saw an article talking about it yesterday but in a completely different context. It was for removing tourists from scenes by taking maybe 15 shots over time (on a tripod) and then using median blending to remove the objects that are different (ie. the people).


I tried doing this last summer at Gooseberry falls in Minnesota. People just stood in one place way too long. Especially those standing in front of the falls waiting for someone to take their picture (I was shooting from the side, up on the wall a bit). I waited and waited...you have to be really patient to make it work! More than 5-8 minutes patient.
 05/29/2013 03:42:02 PM · #8
Originally posted by DrAchoo:
Funny that you bring up median blending because I just saw an article talking about it yesterday but in a completely different context. It was for removing tourists from scenes by taking maybe 15 shots over time (on a tripod) and then using median blending to remove the objects that are different (ie. the people).

The article linked in the OP discusses (and displays an example) of this application of the technique. It really only works to remove moving objects.
 05/29/2013 04:50:02 PM · #9
Originally posted by GeneralE:
Originally posted by DrAchoo:
Funny that you bring up median blending because I just saw an article talking about it yesterday but in a completely different context. It was for removing tourists from scenes by taking maybe 15 shots over time (on a tripod) and then using median blending to remove the objects that are different (ie. the people).

The article linked in the OP discusses (and displays an example) of this application of the technique. It really only works to remove moving objects.


Yep. Never tried it for that, but given the math, it should only work when an object only appears the same spot in one (or perhaps a small fraction of) the median-blended frame count.
For noise reduction, I've always used straight averaging. Never tried median. Averaging should be the most accurate method, by the numbers. It would be interesting to compare the results.
 05/29/2013 04:54:02 PM · #10
Originally posted by kirbic:
For noise reduction, I've always used straight averaging. Never tried median. Averaging should be the most accurate method, by the numbers. It would be interesting to compare the results.

Is that what something like RegiStax uses? Could one use that program on a stack of "regular" (not astro) pictures?
 05/29/2013 05:00:17 PM · #11
Originally posted by GeneralE:
Originally posted by kirbic:
For noise reduction, I've always used straight averaging. Never tried median. Averaging should be the most accurate method, by the numbers. It would be interesting to compare the results.

Is that what something like RegiStax uses? Could one use that program on a stack of "regular" (not astro) pictures?


AFAIK, there's no real reason why Registax couldn't be used for non-astro images. In practice, it might be more cumbersome to use for the "short stacks" we normally use in non-astro work. As far as the math, I believe that Registax supports multiple methods for combining images, but off the top of my head I'm not sure which (in fact, there's nearly nothing left on the top of my head, LOL)

ETA: From what I can see in the Registax 5 Manual (this is a little outdated, Registax 6 was released a while back) there seems to be one basic process with some options for the stacking algorithm. What the math behind the basic algorithm is, I don't know.

Message edited by author 2013-05-29 17:07:18.
 05/29/2013 05:32:41 PM · #12
Well, with my (aged) computer I'd probably have to look for RegiStax 2.0 anyway.
 05/29/2013 06:07:32 PM · #13
Originally posted by kirbic:
Yep. Never tried it for that, but given the math, it should only work when an object only appears the same spot in one (or perhaps a small fraction of) the median-blended frame count.
For noise reduction, I've always used straight averaging. Never tried median. Averaging should be the most accurate method, by the numbers. It would be interesting to compare the results.


Waitaminute. Geek alert. Shouldn't median blending work for moving objects if the background is present in (n/2)+1 images where n is the number of shots? Each pixel is going to be the value of the middle image when the images are aligned by value. Let's say it's a B&W image for simplicity and the background pixel value is 100 while tourists represent values 105, 25, 255 and 80. We took 9 shots. The image set will then look like this:

(25, 80, 100, 100, 100, 100, 100, 105, 255)

The median value is 100 (bolded) and so the pixel will be represented as 100 in the final.

Maybe you were referring to another scenario...

Message edited by author 2013-05-29 18:08:50.
 05/29/2013 06:10:06 PM · #14
Yup, that is what median is supposed to do by my reading. Less of a blending than a selection of the median pixel.
 05/29/2013 06:53:39 PM · #15
is this more or less effective than averaging?
 05/29/2013 07:19:14 PM · #16
Originally posted by LanndonKane:
is this more or less effective than averaging?


For removing objects that are moving it would be much more effective. In the example above the median is 100 which is exactly the value of the background and exactly what you want. For average the value would be 107 which happens to be close, but won't be exact and could show up as ghosting.
 05/29/2013 07:53:55 PM · #17
Originally posted by DrAchoo:
Originally posted by LanndonKane:
is this more or less effective than averaging?


For removing objects that are moving it would be much more effective. In the example above the median is 100 which is exactly the value of the background and exactly what you want. For average the value would be 107 which happens to be close, but won't be exact and could show up as ghosting.


I'm more interested in the noise reduction side
 05/29/2013 08:53:35 PM · #18
The issue of regular averaging is that it will include noise in the mix. If a pixel pops in one frame of an even field, then regular averaging will add it to the mix, median averaging will not use that pixel.

Message edited by author 2013-05-29 20:56:08.
 05/29/2013 08:57:04 PM · #19
Originally posted by DrAchoo:

Waitaminute. Geek alert. Shouldn't median blending work for moving objects if the background is present in (n/2)+1 images where n is the number of shots?


Actually, I was imagining a simpler scenario, but I think your hypothesis is correct. Again, take a grayscale image, and imagine the BG value is 100. Entering the scene between shots at some point is a bright object with an average gray level of 220. If we take 5 frames, and the object is present in 2 shots, the values are (100,100,100,220,220) The median value is 100 (same as BG, object does not appear). If the object is present in three shots, the values will be (100,100,220,220,220) and the median will be 220 (same as object, object is visible). So for odd number of exposures, your hypothesis holds. The object must appear in less than half the exposures, and if it does, it will not appear in the final exposure.
For an even number of shots, say six shots, if the object appears in (6/2)-1 = 2 images, the values will be (100,100,100,100,220,220) and the median is 100 (object does not appear. If the object appears exactly N/2 = 3 times, the values are (100,100,100,220,220,220) and the median is 160 (object appears as a ghost image). If the object appears more than 3 times the object appears in the final image.

So, with the exception of the special case of an even number of exposures, your math is pretty much correct. It's not a very intuitive result, but the median is a rather unintuitive statistic. The

Message edited by author 2013-05-29 21:02:30.
 05/29/2013 09:09:20 PM · #20
Originally posted by BrennanOB:
The issue of regular averaging is that it will include noise in the mix. If a pixel pops in one frame of an even field, then regular averaging will add it to the mix, median averaging will not use that pixel.


For real-world images, adding a "hot" pixel to one of the images will most likely change the value of the pixel in the output image, albeit slightly. Imagine a series of some odd number of images, all comprising of a flat field of somewhat noisy gray with an average value of 128 and a standard deviation of 8. If we introduce one image in the series with a single pixel at a level of 255, that pixel in the final image may well be slightly brighter than the rest, because the next value up the list will be the median.
 05/29/2013 09:33:13 PM · #21
Following all this, imagine a shot of a human against a blue sky. In each frame, I have the model hold her arms in a different "semaphore" position. End result: no arms, weird stub shoulders? Y or N? Jejeje...

Message edited by author 2013-05-29 21:33:29.
 05/29/2013 09:40:47 PM · #22
Originally posted by Bear_Music:
Following all this, imagine a shot of a human against a blue sky. In each frame, I have the model hold her arms in a different "semaphore" position. End result: no arms, weird stub shoulders? Y or N? Jejeje...


I think so! Now, would that violate her right to bear arms? And don't ask why she'd want your arms, I don't know ;-)
Anyhow, getting her to hold the rest of her still enough to make it work would be, well, a bear!
 05/29/2013 09:42:28 PM · #23
I wanted to try the exercise with the "single hot spot in a field of gray" but unfortunately the median stacking option is available only in Ps Extended :-P. I will not be defeated, however, I have Extended at work!
 05/29/2013 11:13:31 PM · #24
Originally posted by kirbic:
I wanted to try the exercise with the "single hot spot in a field of gray" but unfortunately the median stacking option is available only in Ps Extended :-P.

Sheesh that's annoying...
 05/30/2013 12:26:32 AM · #25
What? I got excited about this and then I can't do it? Lame.
Pages:  
Current Server Time: 07/27/2025 02:30:30 AM

Please log in or register to post to the forums.


Home - Challenges - Community - League - Photos - Cameras - Lenses - Learn - Help - Terms of Use - Privacy - Top ^
DPChallenge, and website content and design, Copyright © 2001-2025 Challenging Technologies, LLC.
All digital photo copyrights belong to the photographers and may not be used without permission.
Current Server Time: 07/27/2025 02:30:30 AM EDT.