This is a proof of concept image processor which uses a technique of sacrificing image resolution for light/exposure. Two full stops of light can be gained (hence the name) at the cost of halving the resolution.
This could be helpful for astrophotography for example where saving 2 stops of light could mean the diffence between a 30 sec exposure and a 2 min exposure.
Another use case could be to get very low noise images by shooting at ISO 100 and setting exposure compensation to -2 stops (or more if you boost in post).
You can view the output sample file here.
The best way to set this up is with a virtualenv and installing the dependencies like so:
$ virtualenv venv
$ source venve/bin/activate
$ pip3 install -r requirements.txt
This repo includes a sample NEF RAW file which was shot with exposure compensation set to -2 stops. You can run a test preview like so:
$ ./twostop.py preview images/test.nef
Reading RAW file images/test.nef.
Post processing RAW image preview.
Converting image from array.
Post processing RAW image.
Two stop processing.
You can produce a JPG like so:
$ ./twostop.py process images/test.nef
Reading RAW file images/test.nef.
Post processing RAW image.
Two stop processing.
Rendering to file images/test.jpg.
images/test.nef done.
Done, processed 1 images.
NOTE This only works with RAW images, this will not work with raster images (JPGs/PNGs etc.).
Process one or more RAW image files and output to JPG.
$ ./twostop.py process images/test.nef images/test2.nef --expcomp=1.25
This will produce JPG images at images/test.jpg and images/test2.jpg.
expcomp- Exposure compensation. E.g. 0.25 = -2 stops, 8.0 = +3 stops
Process a RAW image file and show a before/after preview + histogram.
$ ./twostop.py preview images/test.nef --expcomp=1.25
file- The RAW file to process and previewexpcomp- Exposure compensation. E.g. 0.25 = -2 stops, 8.0 = +3 stops
It is really quite simple, we are adding together the luminosity of adjacent pixels to create a brighter image. We are taking 4 pixels, adding together the R G B channels of each pixel and producing a single output. This creates an image at 1/4 the resolution which is 4 times brighter (2 stops).
The below figure might help, 4 pixels becoming 1 ...
+--+--+--+
|10|24| +--+--
+--+--+-- => |66|
| 9|23| +--+-
+--+--+ |
| |
+
Since this is just a proof of concept, we do this by a fixed integer amount making the code simple as possible. Halving the resolution which boosts exposure by 2 stops. You could do this by any amount if you used an intelligent downsampling algorithim.
Additive downsampling is probably a good word for it.
One caveat of this is that it also boosts the contrast of the image. A way to fix this might be to only apply this additive downsampling to the luminosity of the pixel (i.e. do a greyscale conversion), then apply that brightness value to the R G B channels separately.