It is recommended that photometry with DSLR cameras and one shot colour (OSC) astronomical cameras be performed on defocussed images.
What happens if the recommendation is ignored, and in-focus images are used?
A preliminary look at this with a DSLR camera indicates that photometric accuracy (defined as measured target magnitude minus catalogue magnitude for check stars and photometric standard stars) is as good as it is from defocussed images. Errors for some measurements are < 0.01 mag for transformed V (as is the case for me with defocussed images).
I also obtained one set of transformation coefficients from in focus images which were virtually identical to coefficients from defocussed images. The difference was < 0.01 for Tv_Bv.
However, the standard deviation of a set of measures is higher for in-focus images. That means greater scatter of data in a time series.
It would be interesting to see the results of in-focus photometry with a 16 bit OSC astro camera (I don’t have one) because the higher QE and greater well depth could capture more photons, diminish SNR and hence lower the SD.
The main advantage over defocussed images? Accurate placement of stars close to the target within the gap annulus as in the attached image of R Gru.
It is all about sampling the Bayer matrix. If you have a long focus OTA with small CCD pixel sizes your star point-spread-function (psf) will cover enough pixels most likely. If you have a short focal length OTA and large pixels you may only get most of the light in one pixel.
To get the “best” multicolor photometry you need to know and understand your imaging systems stellar psf with typical seeing on you Bayer matrix sensor. SNR plays a role as well.
The AAVSO DSLR Observing Manual states in section 4.7 that the images should be defocussed so that stars are “round and occupy several pixels”, which is open to interpretation. The discussion on lenses and telescopes in section 2.2 deals with everything from short focal length lenses to telescopes and deals mainly with fields of view. It cautions against using in-focus images to capture “very faint” stars because artefacts (discussed in section 5.5) would be introduced, but does not refer to in-focus images with brighter stars.
However, the manual states that in-focus images are appropriate when using telescopes with focal lengths long enough to yield focussed star images with FWHM of 8 pixels or more. If this number refers to images before debayering, the FWHM in debayered images would be 4 pixels.
The AAVSO Guide to CCD/CMOS Photometry (which of course is aimed at users of monochrome cameras) advises the FWHM should be 2 to 3 pixels.
The equipment used to yield the results referred to in my previous post comprised a Canon EOS500D DSLR camera with a pixel size 4.69 micron and a Canon 200mm f/2.8 L lens (no longer available as new) stopped down to f/3.5.
With this equipment the FWHM of in-focus stars in debayered images of both blue and green channels is about 2.5 pixels.
I think you can do a realistic estimate here using Python. Assume a star with a Gaussian 2D profile. Overlay a Bayer pattern on the 2D profile, and sum the flux inside of a given aperture. Then shift the Bayer pattern by, say, 0.1pixel (I’d do it radially from some starting position), and determine the flux again. If you plot the resultant flux as a function of the shift, you should be able to see what effect various fwhm can have on your photometry.
Arne
I would jump at the chance to do something like that if I could, but unfortunately I can’t speak Python.
In my short focal length system driven by a Star Adbenturer Pro mount there is one other consideration, because I average measurements from 10 or more consecutive images to get the final magnitude.
Polar alignment is pretty good but not perfect so there is a slow north or south drift over 10 images. There is also east-west drift due to periodic error of the order of a few pixels and greater than the drift in DEC.
Therefore the centroid of a star image drifts across a few pixels of the Bayer matrix during the capture of 10 or more images. I have graphed this. There is a mention of this sort of thing for very small star images in Section 5.5 of the DSLR Observing Manual
I assume the above behaviour contributes to the error which varies from image to image, but this magnitude error does not correlate in time with the periodic error of the mount.
You can use graph paper! The typical seeing at you obs. and the full-width near the noise-floor and the full-width-half-maximum, two circles and a dot at maximum to get an idea of the number of pixels covered. If you are really good you could easily get an estimate of the integrated flux. You know kinda like first year calculus.
Yes I still have graph paper and a compass and other writing implements.
What tool was used to make the graphics, etc. in the DSLR/CMOS manuals?
Some of the issues from the drift mentioned above by Roy might depend of how the camera firmware/software debayer which involves scaling and interpolation. Many problems can arise from sparse-matrix estimation.
Despite my comment about “jumping at the chance” to try Arne’s suggested modelling if I could, and after looking at Jim’s suggestions, my preference is simply to work with measurements on any available and future images. Zooming in on the seeing disc of one star in one channel of in-focus images clearly shows image to image differences in the pattern of pixels displaying obvious signal. I had assumed, as Jim noted, that the debayering and interpolation processes would contribute to the error budget. I have no idea how that might be modelled. The bottom line is what SNR, precision and accuracy can be achieved in practice.
A color camera sensor has a spec of Pixel Size [µm] = 3.76
Does that mean the total RGGB bayer array is that size or does it mean that each B, G, G, R sensor is 3.76 um?
