Orquid "Phoenix"

Orquid "Phoenix"

by ceinmart

submit your photo


Hall of Fame
View past winners from this year

Please participate in Meta
and help us grow.

Take the 2-minute tour ×
Photography Stack Exchange is a question and answer site for professional, enthusiast and amateur photographers. It's 100% free, no registration required.

The idea comes from the fact that in astronomy (amatorial), it's common to take videos from a telescope with a webcam and convert them into a single, less noisier picture.

The principle is that of oversampling, also used in electronics for analog to digital conversion (what photography is nowadays) to obtain a higher bit resolution from a less accurate hardware using averaging over more samples.

What I'd like to do, is a night picture with a compact camera, so without manual control of shutter time, taking a long video and processing it, obtaining a — hopefully — decent photo, for resolution and noise.

Is this possible? Has anyone tried it?

An alternative could be to take several shots, which will likely have a too short shutter time and will be underexposed and noisy, and use them to make a decent picture. Would this work better?

share|improve this question
add comment

3 Answers

up vote 6 down vote accepted

Yes it is and there are several ways to do that.

The easiest which works with any camera (still or video) is to split the video file into a sequence of stills using software like ffmpeg. Then pass the images to an Exposure Fusion software. While it was not it primary intention, Exposure Fusion works really well to blend images which results in lower noise, increased dynamic-range and greater depth-of-field (depending on the fusion weights).

Should you want more direct controls, it won't be hard to load all these images as layers in a software which supports this concept like Photoshop. Then all you need is to set the correct blending mode. I am guessing something like Average would do it.

As @Russell said, this can also be done from images using Multi-Frame noise-reduction. Sony is big on it and so is Fuji in their CMOS based cameras (F and HS series). They call it Pro Low-Light mode. @mattdm said Pentax does it too but from memory I do not remember which one.

I suspect but didn't try it that you would get very similar results using the Multi-Exposure mode present in most Pentax, Nikon, Olympus ILCs (plus the Canon 1D X and 5D Mark III). This is usually limited to 2-9 images. Note that you must enable Auto Gain on Nikon & Olympus and Auto EV Adjustment on Pentax for this to blend rather than add exposures.

Should you want to do your own Multi-Frame noise-reduction you want to use a proper exposure, not under-exposed images. This is because noise is much higher in dark areas of images. Obviously, if the scene is too dark to be exposed properly then you have to work with what you can get. Adding multiple images to create a brighter one is called Image Stacking and is used for Astrophotography. Given the context in your question, this may be what you were looking for.

share|improve this answer
    
For Pentax, it's every SLR and mirrorless camera from the K-7 on. (Including even the Q.) They call it "multiple exposure with auto ev adjustment". –  mattdm May 26 '12 at 14:45
    
Ah OK. Guess I counted it wrong. That's what I mentioned as Multiple Exposure. Good point about the setting, I'll edit the answer. –  Itai May 26 '12 at 14:51
add comment

Yes, it's possible.

My Sony A77 and a number of recent Sony DSLR's have a mini version of this.

They has a "multi frame noise reduction mode" (MFNR) that takes 6 photos sequentially and combines them. The result on noise reduction in high ISO/low light situations is very significant.

Some comparative images here.

It's really of most value at high ISO's that allow hand held use when a tripod might otherwise be needed. I just tried some 6400 ISO shots in my dining room with a fluorescent fitting lighting it for about 4 metres away from behind me. A single 6400 ISO exposure produced a writhing noise filled mess. Books on the table were so noise obscured that I could not read the title on their spines. In MFNR mode the image was noisy but tolerable and book spines clearly readable. End result was noticeably worse than a D700 would produce in a single 6400 ISO shot.

At low ISO the gains are likely not to be worth the loss of definition which is liable to occur. Many opinions and test photos oin web.

One thing that soon becomes obvious is that they are NOT just averaging the frames. A high ISO shot of eg a headlight illuminated billboard at night will produce a clear enough image of th ebillboard and an improvement in the lower light areas. In some recent tests it appeared that the 2nd frame MAY be used as reference frame to improve when the scene is rapidly changing , but this may be a result of the test circumstances. Whatever is being done applies some intelligence and works better than I'd have expected. But not as well as a D700 :-).

So, in answer to your question, the method works in general principle.
If the subject is relatively static you may be better to use a camera with as fast a frame rate as you can get to improve the megapixels. Bottom end cameras with video capabilities will have small sensors and usually get impressive low ligh vision by dropping the shutter speed. In some cameras in very low light the frame rate will be maintained but the actual picture content may change at eg 3 Hz to acquire enough liught per image. In such cases, using the main sensor on lowest resolution setting may be better. Or buy a Sony A33 :-).

share|improve this answer
    
Pentax does the multi-frame thing too, BTW. –  mattdm May 26 '12 at 12:04
    
I remember coming across a photographer in Los Altos who was doing something like this about ten years ago, but the advantage he had was his partner was a NASA deep space imaging engineer. What they were doing, though, was recording a longterm video capture and then (manually) aligning each image over a period of captured frames, then (with software) stacking the frames until they had a "loss less" print (it was also duplicated on the print side with a custom Epson driver his partner wrote). I got the feeling it wasn't as "simple" as he made it sound, but the 48"x36" prints were beautiful. –  Jared Farrish May 26 '12 at 22:33
add comment

Many years ago I used to work at an animation studio, and I developed a stop-motion capture software for them that used this technique to drastically improve the quality of the animation frames. The camera was just a small standard definition video camera of the kind used for security and monitoring, connected to a PC equipped with a video capture card (note this was years before DSLRs existed). When the animator instructed the software to snap a frame, I would capture a short uncompressed video of the scene, then break it into individual frames and do a simple average of all these frames pixel by pixel, to arrive at a single combined picture. The difference in quality was like night and day.

You can do this using a regular video camera:

  • mount the camera on a tripod, set a fixed exposure and shoot a static scene.

  • To break the video into individual frames you can use ffmpeg (free) or if you have a video editor chances are it can do this too. If given the option, use an uncompressed image format for the individual pictures, the images come already compressed from the video, so no need to degrade them even more. TIFF is a good choice for format. JPEG would be the one to avoid if possible.

  • To combine the pictures you can use Photoshop, GIMP, Paint Shop Pro or any other photo editing app that supports layers with opacity. Start by loading your pictures as layers of a single image. Setting the right opacity to do the averaging is tricky though. If you have just two layers, then the opacity should be set to 100% on the bottom layer and 50% on the top layer. For three images use 100%, 50% and 33%, from bottom to top. For four it would be 100%, 50%, 33% and 25%. I'm sure you get the idea by now. The general formula for the Nth layer's opacity is 100/N. Unfortunately this method can't be used for a lot of pictures, due to the limited resolution of the opacity slider which has only 100 settings. It also gets to be time consuming to setup the layers.

  • A better solution for the average step would be to write a little script to perform the operation. It looks like somebody had the idea already.

I hope this helps!

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.