Retouching of astronomical data

for the production of outreach images

2009-05-20 O.Hainaut

0- Introduction

There are many steps between the data coming out of a camera on a professional telescope and the final image that appears in magazines or on our web sites.

They can be split into three main sections:

1- Preliminary steps

The "raw" dataset is a series of exposures taken through different filters. The filters are needed to rebuild the colour information, as the CCD detectors used work in "black and white" for more flexibility and more sensitivity. The basic filters are R (red), V ("visual", green), B (blue), but many others can be used: UV or IR to expand the "colour" range, or filters with a very narrow transmission centered on the colour emitted by atoms or molecules found in space. The most common one is the red H_alpha, that highlights the Hydrogen gas. In each filter, several images are acquired. This is to increase the total exposure time (and see fainter features in the sky), but also to cover for defects and features that appear in the detector. For instance, we often use CCD mosaics, where gaps are present between the various CCDs. Taking various images at slightly different positions enable us to recover the information lost in the gaps.

Fig 1: a raw image (click for full scale)

The individual frames must be cleaned from all the instrumental and electronic effects that affect them. This is briefly described in another page.

Then, the images taken in a same filter must be registered, aligned and co-added.

Fig 2: Co-added image (click for full scale)

At this stage, the images are still in FITS, the astronomical format, with one FITS file per filter. The FITS file have 16 bits (65536 levels of grey). In order to continue the processing, the images are converted into a Photoshop PSD file, that contains one filter per layer. As the final images must be published with only 8 bits per colour (256 levels in R, G and B for a jpeg image), we must squeeze the full 16 bit dynamic range into only 8 bits. We must also change the response curve of the image so to enhance the faint structure. The change of response curve must be done before the conversion from 16 to 8 bits, otherwise some nasty solarisation effects would appear. A convenient way to do this is to use the powerful FITS Liberator plug-in for Photoshop.

Now, the different filters can be mapped to their colours in Photoshop, and the image can be displayed as a colour image. The result looks horrible. The images still contain many artifacts that need to be cleaned away.

Fig 3. Image converted from FITS to Photoshop, adjusting the dynamic range, but not cleaned. (click for full scale)

2- Cleaning

2.1- The artifacts

While some artifacts are already cleaned at the data reduction stage, many make their way up to the PSD file, and must be cleaned manually or using Photoshop filters. Different types of artifacts are found in the images:
a.- Bad columns and traps: a vertical line that must be cleaned by cloning nearby regions. Careful not to create new, fake stars, or to leave "half stars" in their track.

Bad columns. Most of them are single columns, but it can happen that a group of columns is bad

CCD trap: a dead pixel, near the center of the image, kills the top of its column

b.- Stuff on the detector. Although it should not happen, sometimes a foreign body (like a hair, or dust) can find its way into the CCD housing and ends up on the detector. These must be erased by cloning nearby regions. Clean these out

Foreign stuff on the detector.

c.- Cosmic rays: when a high-energy particle hits the CCD, it appears as a very sharp bright spot on the image. They can usually be identified because they are much sharper than the stars. They must be removed by cloning nearby regions.

Cosmic rays, that appear as bright dots.

Zoom on cosmic rays. They appear as dots or tiny streaks. They are always very sharp.

d.- Saturation bleed: each pixel in the CCD can store only a certain amount of electrons. If a pixel is illuminated by a bright star, that pixel will fill up and start to overflow into neighboring pixels. When the image is read out from the detector, this causes trails that must be removed. Clean out by cloning nearby regions.

Saturation trails. Moderate case on the left, and nastier on the center and right.

When repairing the region right next to the star, it sometimes works best to copy an horizontal strip including the star with the horizontal diffraction spike (see g- below), rotate it by 90deg, and then use this rotated star to patch the original. One must be careful not to hide real stars doing this.


1- An example of saturation bleed correction: the starting point.

3- Rotate 90deg, and align

2- Select an area including the star together its diffraction spikes

4- crop and mask the rotated region so it does not hide anything important.
e.- Oversaturation dot: in extreme cases of saturation, the center of a star can appear black. Clean out by painting over.

At the center of a saturated star, it can happen that a black spot appears.

f.- Satellite streaks: it can happen that an artificial satellite crosses the field of view, leaving a streak. Because the satellite rotates, its brightness varies with time. The intensity of the line can therefore vary. These marks are removed by cloning nearby areas.

The mark of an artificial satellite.

g.- Diffraction pattern: the "spider", that is the arms that support the secondary mirror of the telescope, causes a diffraction pattern that usually appears as a cross around the brightest stars. This pattern is part of the image, and must not be cleaned out. Careful, sometimes the diffraction pattern can be hidden by the saturation trail. The diffraction patterns are part of the image, they should not be removed.

A diffraction pattern (in this case, a cross at 45deg). Not to be confused with the sharper saturation trail (in this case vertical)

h. Background steps: Sometimes, the combination of various images taken at slightly different positions results in steps in the background of the image. This is difficult to correct.

Background steps caused by the co-addition of various images

i.- Ghosts: It can happen that the light of a bright star gets reflected somewhere in the instrument, and forms secondary, out of focus images, that appear as large doughnuts. Sometimes they are left in the final image, sometimes not.

Reflection ghosts.

2.2- Before and after

The different layers of the image must be cleaned individually. Looking at the other layers may help determining what should have been visible under a defect. Here come some examples of "before" and "after".
A small segment of an image, before and after cleaning. Red layer only. Click the images for full size.
A small segment of an image, before and after cleaning, combining all layers. Click the images for full size.

3- Final cleaning and colour balance

After all the image layers are completely cleaned, there is still some work to be done. The various layers must be balanced to best reproduce the appearance of the object. This step involves a combination of scientific (such as the spectrum of the stars, the sensitivity of the camera in the various filters) and artistic inputs (the image must look good!). Also, a final adjustment of the response curve can enhance some details. In some cases, advanced filters can also be applied to further enhance some features.

The final, stunning image is then ready for publication.

The final image. Click for full size.

---oOo---