Henriette and I ran the BBSO linear extrapolation method for removing scattered light with two settings of a parameter that is arbitrarily set. We got output for many files with these two settings and can therefore compare the two results for a single image and measure the difference.

On the image below we see the relative differences between two pairs of two such processed images.

On the left-hand side of the upper image the white cones correspond to percentage differences in the 10-30% range. They gray or black areas are much lower.

The second image is doing better with most differences in the single-% range.

The images processed here are single exposures in CoAdd mode – so there is lots of noise on the DS and the sky. Performance of the BBSO method will probably improve with higher SNR.

We changed the processing parameter (cone width) from 8 degrees to 6 degrees. BBSO uses 5 degrees (in the JGR Qiu et al paper). We can now look at all images where we have these two processing results and make some summary statistics.

Below are two histograms of just that for 383 pairs of BBSO-method corrected images- the first shows the histogram of mean difference values (they are in %). The other is the histogram of the median difference value. Both measures of performance are centred over 0% – the mean difference is a broader distribution than the median – by about a factor of 10 (FWHM mean is 2% while FWHM median is 0.2%). The differences were calculated for all pixels (DS sky and DS itself, but excluding BS and BS sky) corrected by the BBSO method evaluated with the two settings as described above.

The result suggests that correcting the DS for scattered light using the BBSO method and extracting a subregion mean will give unbiased results with a few % scatter, while extracting the median of some corrected subregion will give the answer to a few tenths of a percent.