Return to IPP for PS1

I'm testing the improvement we might expect if we were to apply pre- and post-scan fitted subtractions. I don't really have access to the pre- and post-scan data values, so I've approximated them with a short dark-corrected dark. the dark correction removes some egregiously bad regions from the modeling discussed below.

I've taken a 10s dark image, generated the dark-corrected image, and then corrected each row for a linear trend. To do this in a way which mimics the possible pre- and post-scan correction, I measured the median of the first 16 pixels in the row and the last 16 pixels in the row, then used these two values to set the slope of the bias signal.

The results are quite encouraging. Below, I show a series of plots for each of the cells xy70 - xy77 for chip 12 in a specific image (this was a region which was clearly affected by the strong 'streakies'). The two line plots show the before (black) and after (blue) data value histograms for the cells (one plot is linear, the other is log(Npixels)). The red lines show Gaussian fits to the histograms. The basic sigmas improve significantly (from 8.4 - 10.0 to 6.5 - 7.6). More striking, however, is the improvement of the wings of the distributions. This is most obvious in the log plots, where the corrected histograms much more closely follow the Gaussian model than the uncorrected histograms. The images show the after and before greyscale jpegs (dynamic range of -100 to +100 DN).

Attachments