Briefly, the scattered light subtraction was done by registering the bright and dark images, subtracting the dark from the bright to remove the source, smoothing to remove any residual sources left over after the subtraction, and subtracting this smoothed sky image from the bright frame.
Now for the details, for those familiar with the drizzling algorithm. For each band the steps are as follows:
The large-footprint dark image is subtracted from the large-footprint bright image to remove sources.
Wfixup is used to fill in gaps in this image.
This sky image is block averaged over 3x3 pixels (to speed computation time in the next step).
The sky image is median filtered over 25x25 pixels to remove any residual sources.
The smoothed sky image is block replicated back to the original pixel scale.
The smoothed sky image is subtracted from the small-footprint bright image. Because this image has holes in it where there are no counts, the sky subtraction is only done on pixels where the value in the weight file is non-zero.
The sky-subtracted image is added to the dark image with the appropriate weight.
The final inverse-variance weight file is updated.