Stacking images for Noise reduction.

[jhall]

Recently been doing some longer exposures with my 20D. With respect to the increased

amount of noise, I've done a little research on "stacking exposures". From what I

understand it's a pretty common practice amongst astophotographers. Can

anyone point me to some good tutorials and or step by step explanations of this

procedure?

 

I'm an avid Photoshop users so the methods including this piece of software I'm paticulary

interested in.

[roger_kallet]

Hi J,

 

Try this. Do a Google search for "stacking exposures" within photo.net. This is done at the top of the "post a new message" window.

It seems there are many responses to this topic.

 

Hope this will be helpful. By the way, great picture.

 

Cheers,

 

Roger

[jhall]

Ya, I did that and haven't really found what I'm looking for... Thought someone around

here may have a more tailored response.

[MichaelChang]

The concept is pretty simple, J. <br><br>

The idea is to layer many exposures to effectively increase Signal-to-Noise ratio. When you layer several shot of the same scene, any fixed object present (usually the desired signal) will be directly added while random signals (usually noise) will be averaged. <br><br>

 

Here's how you'd stack: Say you have 10 shots of a night scene. Pull up the first image as the base layer, then add the second at 50% opacity, the third at 33% opacity, fourth at 25%, etc.. In other words, 1/2 of the second, 1/3 of the third and so on. They must be of the same exposure meaning you don't stack images from different camera settings. At some point you'll reach diminishing returns which from my experience is around 8 images (we're talking common photography) - the next increment would be 16 images to gain an additional 3dB noise reduction. <br><br>

Many of my older (SONY DSC-F505V) night shots in my portfolio are done this way, and I still do it if necessary even with modern cameras (Nikon D70) with better intrinsic noise performance.

[jhall]

Thanks you michael. I suspected most of what you explained. The one gleaming point I've

missed is that each stacked image is 1/2 the other. This all makes great sense now.

Thankyou.

[awindsor]

If you are using windows then there is a free piece of software called ImageStacker

from Max Lyon's of Panorama fame (the wide images that is not the TV program).

<P>

 

It is called <A HREF="http://www.tawbaware.com/imgstack.htm">ImageStacker</

A>.

[Ed_Ingold]

That's essentially what the new HDR Merge function does in Photoshop. If you have a difficult scene, like with important detail in both sunlight and shadow, you can take two or more frames and merge them into a 32-bit/channel file. You then use the usual tools - Levels, Curves, Light/Shadow - to adjust the image.

 

I have a D2h, which is rather noisy in the shadows. The use of HDR nearly eliminates the noise you would get my merely adjusting the shadows.

 

I don't have a book other than the manual that explains HDR, and I didn't even need that. It's relatively intuitive to use.

[awindsor]

HDR is normally used when you have two or more images with different exposure levels and they are not blended uniformly; the lighter exposures are used to extract shadow information while the darker exposures are used to obtain the highlight information. I do this with curves and blending options on layers (still using PS7).

 

Image stacking is typically done with images of equal exposure vlue and they are simply added or averaged (typically if you are simulating an exposure of hours you break it up into shorter exposures which you linearly add). Each exposure should be long enough to You can average several identical shots taken at the same time to reduce noise. This only works if the camera and subject are completely stationary. This is what multipass does on a scanner.

[sean de merchant httpw]

HDR requires having variations in exposure. If all the shots have the same or very similar exposures, then Merge To HDR in CS will refuse to do so due to the lack of dynamic range.

 

Albeit, if you bracket properly, Merge To HDR will yield a nice noise free image.

 

Stacking using layer opacity (50%, 33%, 25%, 20%, 13%, ...) works well and will remove a lot of noise (if not all) and add spacial resolution at 5 frames with a digicam (noisier small sensor). This should be adequate to your goals.

 

If you want to try astronomical tools, take a look at Registax (free) from http://registax.astronomy.net/ and make sure you have a fast system with adequate RAM (CPU matters a lot for this type of calculation, especially if inputting 16-bit TIFFs). Read the manual which is available on the download page to get a grasp of things. Albeit, this tool has many features not applicable to your problem (tracking, ...). Albeit, you could use it to convert an AVI from a web-cam into a moderate resolution image with decent detail.

 

hope this helps,

 

Sean

[Ed_Ingold]

I omitted the following from my original post. Using the HDR tool, I take two or more images of the same scene, exposing for different areas.

 

Simply "stacking" images with the same exposure would reduce random noise through averaging, much like multiple-scan film images. The improvement in noise would be proportional to the square root of the number of images stacked.

 

Stacking in this way would do little to improve the useful dynamic range, since shadow areas have so much more noise when underexposed than if you expose for the shadows directly. I expose perhaps 2 to 4 stops more in the "shadow" image than for the "sunlight" image, which effectively boosts the dynamic range by 2 to 4 stops.

[MichaelChang]

"Stacking in this way would do little to improve the useful dynamic range, since shadow areas have so much more noise when underexposed than if you expose for the shadows directly."<br><br>

Edward, doesn't an increase in Signal-to-Noise ratio directly translate into improved Dynamic Range (the ratio of highest to lowest signal levels)?

[sean de merchant httpw]

While simple stacking may not improve dynamic range, living in the Pacific Northwest of the USA where it is often moderately humid or cloudy I find that ambient lighting conditions do not have enough contrast to use the full contrast range of a RAW file. Hence, simple stacking whould work okay most of the time.

 

Albeit, I have also tried HDR in place of simple stacking using manual exposure to get highlights down to the 3/4 tone and the deepest shadows above the 1/4 tone have gotten nice noise free results.

 

The biggest trade off though is needing to touch the camera to get more than 5 stops of dynamic range. With simple stacking or bracking by +-2 stops I can shoot the full set without ever touching the camera by using a remote shutter release. To get more dynamic range one must manually change camera settings by touching the camera unless your camera is tethered to a PC. This touching the camera can often cause misregistration of frames and having to get the individual shots registered (aligned) before doing HDR can create more work. I have yet to do a comparison to compare the quality of shadow detail from the two methods. I do know they get comparable results in midtones and highlights.

 

And using HDR and failing to have all your frames in register induces blurriness in the output which looks like noise with a very strange character.

 

some more thoughts,

 

Sean

[sean de merchant httpw]

"Edward, doesn't an increase in Signal-to-Noise ratio directly translate into improved Dynamic Range (the ratio of highest to lowest signal levels)?"

 

I believe Edward was implying that the S/N ratio in shadow details is lower in the HDR image than the stacked image. Hence, a lower S/N ratio in the shadows would create a finer lowest value that can be resolved against background noise which implies having a greater dynamic range.

 

Whether this assertion is true or not is likely to be image content dependent. In a low contrast scene one could easily capture all detail above the mid-tone without blown highlights which would likely mitigate the advantages of HDR.

 

enjoy,

 

Sean

[Ed_Ingold]

I stand by my statement that the dynamic range is not significantly improved by stacking or averaging images. Averaging makes the noise less blotchy, hence less objectionable. However, the incremental levels are very close together at the black end of the scale, due to the linear nature of digital imaging. If noise is truely random, a stack of 4 would improve the s/n ratio by a factor of 2, or 1 stop. However, we are still talking about black and nearly black - as dark as you can get.

 

The signal to noise ratio is higher (higher is better) if the exposure is increased for the shadows than if the shadow area is simply lifted in processing, assuming the ISO level is constant. Noise is hardly noticeable above the middle of the exposure scale (Zone 5). What appears as nearly black (nearly zero) in an image exposed for the bright areas becomes grey (mid-scale) in an image exposed for the shadows. That is captured in the HDR merge.

 

HDR Merge acts to compress the dynamic range of the scene (much like negative film) so that it can be represented in a print or image. A raw HDR image looks pretty much like the original scene - sunlight and dark shadows. The difference is that the shadows can be lifted in post-processing AND SHOW DETAIL, with very little noise. That is the difference!

[gooseberry]

<p>Just in case someone's still looking for the answer:</p>

<ul>

<li>extended PS version @ <a href="http://www.digitalphotopro.com/technique/revolution/reduce-noise-with-multiple-shots.html">http://www.digitalphotopro.com/technique/revolution/reduce-noise-with-multiple-shots.html</a></li>

<li>a tad more laborious process @ <a href="http://www.tipsquirrel.com/index.php/2010/06/noise-reduction-with-smart-object-stack-modes/">http://www.tipsquirrel.com/index.php/2010/06/noise-reduction-with-smart-object-stack-modes/</a></li>

</ul>

[howard_grill]

<p>Thanks Tomek...I guess you never know when someone will Google and come back to an old thread. Thanks for the update...very helpful!</p>

[gooseberry]

<p>Exactly my thinking, Howard :D<br /> To this end, I'll chuck in a couple more references:</p>

<ul>

<li><a href="http://www.cambridgeincolour.com/tutorials/image-averaging-noise.htm"><em>Noise Reduction by Image Averaging</em></a> article on Cambridge in Colour; and</li>

<li>Ron Bigelow's <a href="http://www.ronbigelow.com/videos/noise-avg/video/noise-averaging.html"><em>Noise Averaging</em></a> video tutorial for those who prefer watching a demo over reading.</li>

</ul>

<p><br />Now, the funny thing is, it came hand to me earlier today: I wanted to do just that, but wasn't sure I remembered correctly every step of the procedure, so instead of searching for it on the web, I knew where I'd find the links (must be good karma, ha-ha! ;-)<br /> Here's <a href="../photo/11912770&size=lg"><strong>the result</strong></a> of my application of this technique (actually, it's quite heavily processed shot, but that was the starting point.)</p>