Rejection versus acceptance of poor quality deep sky sub-exposures
Introduction: Conventional wisdom states that deep sky sub-exposures with distorted stars due to e.g. guiding problems, flexure, wind, etc. should be discarded and only ‘good’ images used for stacking. I find that such a rejection does very little to improve the final image quality – in fact the limited S/N ratio when using few sub-exposures is a much worse problem.
Setup: Borg 100ED f=640mm refractor, FLI DF2 focusser, SBIG CFW8 filter wheel, SBIG ST10XE camera (2.2 arcsec/pixel), Takahashi EM-10 mount.
Image aquisition: Target was IC443 through a H-alpha filter. My guiding performance was quite poor, I suspect due the use of a very faint guide star (mag. 8.5, 20sec guide exposure time, typical star adu=800). A total of 18 images each with 900sec exposure time were aqcuired.
Image processing and analysis: I performed the usual basic callibration (darkfield subtraction, flat fielding). The images were then manually inspected for ‘star image quality’ and sorted into four categories. Central region closeups of the images are shown below along with measured stellar FWHM and aspect ratio (percentage difference between star diameter along elongation direction) using CCDInspector:
Best images: 3,5,6,13,18;
(‘best’ – average FWHM = 3.1, average aspect = 20%)
Medium images: 1,2,7,8,9,12,16,19;
(‘medium’ – average FWHM = 3.1, average aspect = 37%)
Poor images: 2a,10 ,17;
(‘poor’ – average FWHM = 3.8, average aspect = 34%)
Very poor images: 14,15;
(‘very poor’ – measurements of CCDInspector cannot be trusted due to severe distortion)
The precise categorization of a single image is always subject to discussion, but the general trend from ‘good’ to ‘very poor’ should be fairly clear. Another way of visualizing the categorization is done in the plot below:
Each dot corresponds to an image (labelled). The closer an image is to (0,0) the rounder, smaller stars it has. As you can see my initial categorization is pretty good. The only thing that jumps at the eye is to switch category on image 7 and 18, but doing so will not affect my general conclusion.
I will now try registering and stacking (using Registar 1.0, median/average combine) these images in the following manner: best, best + medium, best+medium+poor. Closeup of a central region is shown below for each resulting image:
As expected the S/N ratio increases with number of images used for the stack. However, it is – to me, at least – very surprising that the star roundness / image sharpness is very nearly the same. To quantify this I measured the stellar FWHM and aspect ratio of the image crop above:
FWHM (pixels) aspect %
best: 3.10 11
best+medium: 3.10 23
best+medium+poor: 3.02 19
No significant differences can be seen in FWHM and aspect ratio compared to the single images in the ‘best’ category! In fact, the aspect% of stacked images is lower than the group averages. Using only the best images does produce slightly rounder stars (11% aspect ratio vs. ~20%), but the reduced S/N ratio dominates the image.
Conclusion: For the images used here careful filtering of sub-exposures according to star quality does not yield significant improvement in the stars of stacked image. In fact, the S/N improvement when using most of the sub-exposures is a much more prominent effect. For optimal processing one could use the three stacked images selectively according to region brightness, i.e. ‘best’=brightest regions (stars), ‘best+medium’=mid-level brightness regions (weak stars, main nebulosity) and ‘best+medium+poor’ for the weakest regions such as background and very faint nebulosity. That way most of my hard earned data will be used while displaying only the best from each image stack. Does this conclusion depend on what images are used and the kinds of distortions they suffer from? I don’t know, but I have seen this effect before during my deep sky work. I am very interested in hearing about YOUR experience of this!
Feb. 20, 2010 – Mikael Svalgaard