Affiliation
American Association of Variable Star Observers (AAVSO)
Sun, 02/09/2020 - 11:05
In A Practical Guide to Exoplanet Observing images are shown as being saved in 32 bit format when output from DP. Are there specific reasons why you want to save them in this format? It doubles the size of each image file. I use virtual stack when importing sequences into AIJ so there is no problem in that respect but my internet service is DSL and it double the time to upload images to the cloud to share with a research team from a day and a half to 3 days!
No jokes about my internet please. There is no better alternative where I live and I would be happy if Elon Musk and the other defilers of the night sky never launch enough satellites in space to provide service.
Brad walter
Hello Brad,
that is very common practice, I'd say that every software made for correct image processing/analysis does that. When you have raw frames, pixel signal levels are given (typically, but not always) as unsigned two-byte integers, giving digital signal range 0-65535, that is sufficient to represent typical pixel full well signal of CCD cameras. Integer math (subtracting one integer from another) lives well within that range and in signed int data type...
However, when one divides (e.g. for flatfielding) raw object frame with flat frame, the result is a fraction and you need a floating point type data format to describe it. Smallest useful one of them I'm aware of is 32-bit float. In principle, if one discards fractional part, data could still be saved in integer format, but with the expense of loss of accuracy/+precision.
You could try to compress your FITS images either using some ordinary algorithm (e.g. zip) OR FITS lossless compression algorithm (fpack/funpack). The gain is usually ~2x...
Best wishes,
Tõnis
I asked a similar question on a different forum and this is the response I got from Dr. Karen Collins:
Is there an overwhelming need/benefit to save the CCD data processor in AIJ in 32 bits as opposed to 16 bits? Does it affect the photometry at all? (I archive all my observations and I'm trying to save hard disk space).
For bright stars, it probably makes no difference. For faint stars saving in 16 bit might add noticeable noise, but I've never specifically tested it. As a note, if you are running photometry in data processor (DP), all of the photometric measurements are taken from the 32-bit images before saving in whatever format is specified. If you are just calibrating using DP, and then loading the stack of saved images back into AIJ, then of course the previous comment doesn't apply. I hesitate to mention it because I can't provide the support to make it work on your system, but another option would be to use FPACK to compress the images (either 16 or 32 bit) after they are saved to disk by DP. You can read more about it here: https://heasarc.gsfc.nasa.gov/fitsio/fpack (edited)
fpack Home Page