Low light photography

Soldato
Joined
13 Jan 2003
Posts
24,365
I normally do astrophotography but I have been attempting to use the same techniques for low light static subject photography without using a flash.

Astrophotography actually uses a set of photographs that are then aligned so they can be "stacked". This stacking acts to reinforce the signal whilst the noise that shifts (CCD noise for example) does not reinforce.

In this shot, the subject is illuminated and is held behind a glass cabinet so no flash can be used. The additional lighting in the area also reflects on the outside of the cabinet to create reflections of anything around it - including people looking at the subject...

Here is a crop of one of the "subs" one of the single sub-images of the ten taken, note the light reflection and the person looking also being reflected:
IMG_2309_Preview01.png


Now here's the final image resized to 588x784 for the forum.. you'll see there's virtually no reflections, a large reduction in JPEG artefacts and CCD noise.
integration_588x784.png


However here's a crop of a single coin showing the clarity:
integration_Preview01.png


Whilst stacking I used Windsor sigma noise rejection which rejects low and high level noise but also rejects movement between frames as noise and here's the high rejection. Note the reflections being removed as noise:
rejection_high_resized.png


Cool or what :D
 
Last edited:
The new DSLR I have does MFNR (Mutli-frame noise reduction) in the camera, using 6 frames that are effectively stacked, that very simplistic mechanism has quite a profound effect on noise in low light photography, but doesn't quite match up to your excellent ten shot example!
 
I forgot to say I was using an old Cannon IXUS 80. 8M pixel, simply using the customer timer with zero time and 10 shots configured - handheld with auto ISO... each sub is supposedly ISO 200 f/2.8 1/8th second exposure stored as JPEGs. Each exposure is only 2.4MB.. but after processing at 64bit pixel depth during stacking the image is ~200MB :D
 
Last edited:
Lets take a simple image and slice though the stack, below, where the target on each sub is 100% aligned (called registration):

stacking.jpg


In this you can see each sub-image (sub 1-6) being stacked.

When stacking noise varies between frames can be identified (green & cyan) also noise that follows a motion (blue) can be detected and removed.

More advanced...

Registration is of target objects can occur regardless of the position or the resolution of the sub-images. With this knowledge we can align the target features in the photos in specific ways to do some other interesting things:

* removing glass noise by artificially moving the images to identify shifting reflections
* enhancing resolution by registering the shifting pixels and mapping onto a higher resolution. This recovers additional resolution.

The Process

The process is simple.

1. Take a set of exposures - these should be the same length, but the subject can move in the sub-image. Actually this works better with a little shift between each frame. The clearer each sub-image the better although slight movement and focus differences will be averaged out.

2. sub-image registration - this is where the common features shared between each sub-image are located and any image transform that needs to be applied to align them is discovered.

3. Integration - this is where each registrated sub-image is stacked to produce a final image. The summed RGB values can be averaged over the number of frames. With additional forms of processing allow variations in the frames (sigma noise) to be identified.
 
Last edited:
Another using the same method, although this one can't use sigma noise rejection because of the water mist nozzle near the light at the back:
integration_1024x784.png


The sub-image crop of a specific area:
sub_crop.png


The final integration crop of the same area:
integration_crop.png
 
Last edited:
stone_1024x784.png


Registration points being the dark dents in the rock. One point that I think would improve this - stopping down will give a larger depth of vision which should help the focusing.
 
Last edited:
That's excellent and I never thought of using such a technique for that situation. Given the number of museums and castles I visit I'm looking forward to putting this to use. Cheers :D
 
Last one.. as I suspect everyone is board lol. Roman lead, inscribed with a curse and sat behind glass:

curse_1024x784.png


Native resolution crop:

curse_native_crop.png
 
Back
Top Bottom