Having just spent an entire Saturday working on an image, I thought I would give a synopsis of what it takes to create a deep space image. Many people are not aware of the details, patience, statistics, and computer processing power that is required. Many people think that you just hold a camera up to an eyepiece and take a snap. Unfortunately, it is more complicated than that!
During the processing of my last image, I set up Ffmpeg to capture my desktop at one frame a second and create a video at thirty frames per second. This shows the entire routine of processing all of the data I collected into a single image. This video is 22 minutes and 52 seconds long, so for thirty frames per second, at a second of real time per frame, that means it took me and my poor computer almost 11.5 hours to create the image of NGC 7640. I bet this old silicon is happy to have had a new heatsink recently!
Step One: Acquisition
The first thing you have to do with the data is collect it! Thankfully, I’ve spent a lot of time automating my telescopes, and this allows me to tell them what to do, what to image, and when, all while I am asleep. I pick targets using Stellarium, where I have recreated the view of the sky from my telescope, including obstructions such as trees and buildings. This lets me pick a target that is just starting to rise at dusk, and also allows me to pick multiple targets when the nights are long in winter.
The telescopes then take the same pictures of the same targets, over and over and over. I typically like to get at least 30 hours of exposure on a single target. Because of the bad light pollution in my skies, you need more exposure to average out the noise from the light pollution. My big telescope takes four minute exposures and has exceptional guiding, allowing me to keep all the data except for the worst atmospheric conditions. My little telescope takes three minute exposures, and with enough fiddling, I have tamed the guiding to a point where I no longer have to trash exposures.
Step Two: Pre-processing
The steps I take for pre-processing are:
- Calibrate the flat frames by subtracting a master bias frame.
- Integrate the flat frames.
- Manually inspect every single image, separating them by the side of the pier the telescope is on. This is necessary because my Newtonian telescope is poor at holding collimation, and doing flats on each side of the meridian leads to sufficient image correction.
- Calibrate the separated light frames using a master dark and the appropriate master flat.
- Debayer the light frames.
- Weigh the light frames to determine the best frames. I take a handful of the best frames near meridian to create a normalization reference.
- Apply automatic background removal with linear fit to the normalization frames.
- Register and integrate the normalization frames to create a normalization reference.
- Register all light frames to the normalization reference.
- Apply local normalization to all light frames using the normalization reference.
- Integrate all light frames into the final master light.
Step Three: Post-processing
Post processing involves making the picture pretty!
- Remove the background gradients from the image, either using dynamic background removal if they are gentle, or by removing the stars from the image and creating a gradient image to subtract by removing multiscale wavelet layers until only large scale gradients are selected.
- Color calibrate the image using spectrophotometric color calibration, which looks at the flux of stars in each channel and compares them to reference data to white balance the image.
- Deconvolve the image using BlurXterminator.
- Fix clipped stars using the repaired HSV separation script.
- Separate the stars from the image using StarXterminator.
- Stretch the starless image using careful application(s) of generalized hyperbolic stretch.
- Denoise the starless image using NoiseXterminator.
- Contrast touches, such as local histogram equalization or a multiscale approach.
- Saturation boost to the starless image.
- Stretch the star image using hyperbolic arcsin stretch.
- Reduce star size by using several iterations of morphological selection.
- Remove magenta star colors by inverting the star image, removing green with SCNR, and inverting the image again.
- Saturation boost to the star image.
- Combine the starless and star images into the final image.
