Average multiple images, using geometric median. Averaging is done pixelwise, using iterative vaiation of the Weiszfeld's algorithm.

geometric_median.py [-h] [-g] [-p] [-t N] [-i N] [--bias BIAS] [-n ORDER] [-c NAME] [-o OUTPUT] IMAGE [IMAGE ...]

• -h, --help show this help message and exit
• -g, --glob Apply glob on given paths (useful on windows)
• -p, --preload Preload all images to memory. Faster, but requires a lot of memory.
• -t N, --threads N Use N worker threads, default is processor number+1
• -i N, --iterations N Number of iterations of the algorithm
• --bias BIAS Bias of the algorithm, small positive floating-point value. Default is 1e-6.
• -n ORDER, --norm ORDER Order of the norm function to calculate distances. Possible values are 1, 2, inf, -inf
• -c NAME, --colormap NAME Apply specified color mapping before averaging images. Possible values are: none (use RGB as is), projective[:luma-weight=0.3] (suppress importance of luminance by luma-weight factor, value between 0 and 1)
• -o OUTPUT, --output OUTPUT Output image file

Code written in Python 3, but should be easily adaptable to Python 2. Requires following Python modules:

• Numpy: for numeric calculations
• PIL (or its fork Pillow): for reading and writing image files.

Try reduce number of worker threads (default is number of processors).