Are Hubble Images Real? Part I: Image Artifacts

TLDR: Yes, Hubble images are real. This series of posts is dedicated to the scrutiny of Hubble imagery and a broader discussion of the veracity of astronomical imagery. In this post, we’ll look at the distracting image artifacts that are removed when creating press imagery.

Ghosts, spikes, and dragon’s breath…No, it’s not a Hubble-themed spinoff of Game of Thrones (but what would THAT be like?)…These are terms used to describe some of the many artifacts present in the raw data taken by the Hubble Space Telescope. If left in the final image, these artifacts could easily become a source of confusion or distract from the science inherent in the image. It is the job of an image processor to reduce the presence of these image by-products as much as possible while maintaining the integrity of the data in the process of creating press-quality imagery. If you’ve followed along on the blog for any length of time, you’ve likely encountered posts where we have discussed the process of data calibration, pre-processing, and cleaning of the image. These terms describe the ways in which we remove image artifacts. Many artifacts are caught and cleaned up during the pre-processing phase of data preparation. The data are in their most raw state during this phase, and individual images are added together to create the initial set of images that are combined in color to produce a press release image. But before we dig into that process, and in order to better understand the source of image artifacts, we must first take a step back and gain some perspective on how Hubble works. Understanding this will help in identifying some of these image artifacts. Let’s take a look at the path of light entering the telescope before it nestles down into one of Hubble’s detectors and is recorded in an image exposure. NASA provides lots of great resources for those curious about this, but I will summarize it here.

As you can see in the above diagrams, light enters the main baffle and is reflected off of the primary mirror, back through the main baffle to the secondary mirror and then into the central baffle where it is then “picked off” by small assemblies of mirrors within each individual science instrument behind the primary mirror. Of course, it doesn’t end there! See below a diagram showing the internal structural components that make up Hubble’s Wide Field Camera 3 (WFC3) detector, as well as a simplified schematic showing the light path through the instrument.

Even within the detector there are multiple mirrors guiding the light through the instrument and filters and ultimately down to the detector that records that light. The light from distant stars and galaxies that has traveled across space and time for millions to billions of years relatively unscathed takes quite a roundabout path before being recorded in an image! It is no wonder then that within that path, there are multiple areas where unwanted light from reflections within this system can also make its way into the image.

Now that we have a better understanding of how light makes its way through this complicated system, let’s define some of the terms we use to describe the image artifacts, along with examples for each, and strategies we use to mitigate them. I have separated this list into broad categories intended to indicate the origin of each type of artifact.

Artifacts that originate within the detector itself

Charge Bleed (saturation or bloom)

Charge bleed occurs whenever there is a source in the field of view that is so bright that it overwhelms individual pixels within the detector. Hubble’s imaging detectors are comprised of charge-coupled devices (CCD), the same technology that allows you to take pictures on your cell phone. Think of the detector’s individual pixels as an array of buckets, and the photons of light that are detected as rain drops falling into those buckets. To create an image, those buckets have to collect rain for a certain amount of time and then transfer the ‘image’ off of the array and begin taking a new picture. The transfer process in CCDs is often described as a “bucket-brigade” as the collected photons are transferred from one bucket to the next until they reach something called the “framestore” which temporarily holds the complete image as it is stored in memory. If, during the creation of a single frame, there’s an area where rain is falling at a higher rate, those buckets will fill up faster. If they fill up too fast, they may begin to overflow into neighboring buckets, and will quickly fill those buckets as the image is transferred to the framestore. In images with very bright stars, this can be very difficult to clean up. However, if there is a mostly uniform background behind the bright star, cleaning this artifact should be relatively straightforward. The most desirable method is to use real data from another image of the same source rotated such that the data in the bleed can be substituted with data from another image. In the absence of that ideal situation, charge bleed can be repaired by selecting and copying a clean area of the star perpendicular to the bleed, rotating it ninety degrees and cleanly blending it into the charge bleed, being sure to not obscure or add elements to the image.

Charge Transfer Efficiency (CTE)

Continuing with our bucket analogy from above, imagine that some of the buckets are unable to transfer all of their contents during the transfer period but instead dribble it out over time. This would have an effect on the efficiency of the bucket to transfer all of its charge to the next bucket at one time. The resulting image ends up showing streaks where those photons escaped their buckets between transfers. CTE does change over time, and it is unavoidable that the CCDs degrade and their charge transfer becomes less efficient with age. The software wizards at STScI have built tools to correct for this effect into the data calibration software that essentially packs that charge back into its bucket. This repair is applied as a part of the standard processing pipeline, so the end user downloading data from MAST will almost never encounter this effect unless specifically looking for it.

Chip Gap

This artifact is well known, as it is a part of the design of the detector. More than one CCD can be combined together to create a larger field of view for the detector. Unfortunately, this means that there is a gap between the different CCDs. Both the Advanced Camera for Surveys Wide Field Camera (ACS/WFC) and the Wide Field Camera 3 (WFC3) UVIS detector have chip gaps. Thankfully, observations can be carefully planned so that the detector is moved between exposures, filling in the areas where there is no data. If the image happens to be a single frame, and there is no data to fill in the gap, other data, either from other Hubble observations, or even sometimes ground-based observations can be used to fill in the missing data. If real data cannot be found from any other source, the image can be cropped to avoid showing the gap, or as a last resort, the gap can be filled with the average background level so it doesn’t stand out as much.

Pixel problems: Hot, dead, unstable, cold, bad pixels

These are all ways of describing individual pixels in the detector that do not perform as they were intended. In some cases, they are stuck on and always produce a maximum signal (hot pixel), and in others they do not respond at all to incoming stimuli (dead). These are usually well known, mapped out on the detector, and can be handled during data calibration. For example, there is a well-known circular area of dead pixels within the WFC3/IR detector that has been lovingly nicknamed the “death star” and is even included in planning software to help scientists avoid placing anything important on that part of the detector when planning observations. These problem pixels are unavoidable in the manufacturing process of imaging detectors and not usually a sign of something worse. However, the detectors do degrade over time in the harsh conditions of space, but again, these problem areas can be accounted for and handled in calibration. Taking multiple exposures and slightly moving the detector/telescope between exposures also helps to mitigate these problems area once the data are combined together to create an integrated image.

Detector Field of View

Similar to the chip-gap artifact, the design of the different detectors can impart its own unique signature on the composition of an image. Of course, the most iconic example of this is the unique stair-step look of Hubble’s now decommissioned Wide Field Planetary Camera 2 (WFPC2) detector. This pattern was the result of the detector being designed to have four CCDs arranged in a square, with one of those CCDs having double the density of pixels compared to the others. This was the “Planetary Camera” or “PC” chip, so named because it was designed to produce higher-resolution imagery of our solar system’s planets. When an image using all four CCDs is created, the doubled density of the PC chip effectively shrinks the image to half the size of the others, thus creating the stair step. The only way to avoid an awkward field of view is either by combining exposures to create a square or rectangular mosaic, or simply cropping the image. Of course, plenty of iconic Hubble imagery from the WFPC2 era simply left the stair-step in place!

Mosaic Artifacts

These artifacts really stand out as their own category, but I’ll group them here with detector artifacts for simplicity. As we’ve covered in previous posts, Hubble’s overall field of view on the sky is very small. In order to make images that cover a larger portion of the sky, astronomers plan out multiple overlapping observations to create a mosaic. These data then need to be carefully calibrated so that the overall background levels match from one frame of the mosaic to the next. Even with such careful calibration, there will always be areas near the overlaps that don’t quite match and will need to be dealt with, usually through the use of local brightness/contrast adjustments to match the background levels. Sometimes, science programs are less concerned about creating a perfectly square mosaic, and so there may be gaps in the data, those areas can be filled in with data from other programs, or with ground-based data if available. If not, we can fill in small gaps with the average local background level.

Artifacts introduced by reflections within the optical system

Diffraction Spikes

Of all the artifacts discussed thus far, and indeed, still to be discussed, diffraction spikes are unique. Of course, they are an image artifact, but they have become synonymous with our concept of what a star should look like, so much so that they are often intentionally left in an image. We interpret them as a pleasing element in the image, and they add a sharpness to the overall quality of the image. Removing them, especially from an image filled with bright stars would most likely be a futile endeavor.

These spikes are actually a result of the design of the telescope and are caused by light bending around the struts that support the secondary mirror within the main baffle of the telescope assembly (think back to the first figure in this post showing Hubble’s optical telescope assembly). Actually, we often see this same phenomenon in terrestrial photography—caused by a camera’s optical system. The blades that make up a camera’s iris diaphragm, when stopped down to very high f-numbers, or a very small aperture, tend to produce a similar effect on bright points of light. This is generally seen as a pleasing effect, and is often intentionally planned to add a dynamic element to the photograph. Back to astronomical imagery, it is also interesting that different wavelengths of light scatter around the struts differently. This is what gives the diffraction spikes in color images their subtle rainbow look, which also adds to the image aesthetics.

Although I’ve clearly made the case that we usually keep diffraction spikes in images, there have been examples where they have been removed because they may distract too much from the science story of the image. One such example is a 2019 press release image of Eta Carinae. The rays of light that appear to shine through the homunculus (the dumbbell-shaped nebula at the center of the image) and illuminate the surrounding red gas are one very important element of the image. In a chance alignment, the strong beams shining towards the lower right of the image just happened to overlap with diffraction spikes from the bright central star. Removing those spikes aided in the telling of this science story.

Donuts, Ghosts, and Glints

Any bright star in the field of view of the detector (and even just outside of it) is going to produce reflection artifacts. The location of these artifacts depends on how far “off-axis” the bright star is. If the star falls near the center of the detector, there will be a large glowing ghost of its reflected light nearby. The farther away from the center a star is (off-axis), the more offset and distorted these reflections become. Some take the form of donuts, or figure 8s, and there can be multiple reflections from a single bright star within the field. What’s more, bright stars that aren’t in the field of view of the detector, but still reflected within the optical system (remember, the instruments only pick off a small amount of the light actually entering the telescope), can produce glints that wind up on the detector. There is one type of glint in particular that has come to be known as “dragon’s breath” for its unique, sometimes quite large, glowing signature on the image resembling a torrent of flame coming from a dragon. The image processor should do everything within their power to reduce, remove, or mitigate these effects to reduce the potential for confusion in the final image. This could involve replacing the affected area with data from other observations, and/or using tools to reduce the brightness of the reflection. A last resort option, depending on the severity of effect, and where it falls in an image, is to replace those pixels with the average background level.

Point-Spread Function (PSF)

The PSF is another by-product of the optical system of the telescope. We must remember that even with Hubble’s exquisite angular resolution (its ability to actually resolve features on very distant objects), it is not actually resolving the surfaces of distant stars. All of the stars that Hubble can see are, in fact, pinpoints of light, but the brighter a star is, the more its light spreads out within the detector—this is what is meant by point spread. Hubble is not seeing the surface of those bright stars, it’s just that the detector system is overwhelmed by their brightness, and each detector and filter combination imparts a unique look to that spread of light. This is why very bright stars are surrounded by colorful halos of light in color images. Excessively large PSFs that distract from their surroundings can be handled through local brightness or contrast adjustments when compositing the image.

Artifacts from outside sources

Cosmic Ray

Cosmic rays are a fact of life when you’re an orbiting observatory just trying to snap some nice pictures of space! Most of these particles are absorbed by Earth’s atmosphere and magnetic field, but Hubble operates far enough above the atmosphere to be constantly bombarded by a mist of high-energy particles zipping through space in all directions. Cosmic rays are energetic enough to pierce through Hubble’s exterior, and the ones that make it down to the detectors impart a charge on them, effectively saturating individual pixels while an exposure is being made. Depending on the angle at which the particle strikes the detector, it will appear as either a few pixels, or a long streak. The good thing about them is that they are short lived, constantly changing from one exposure to the next. Combining a few exposures together and using software designed to identify cosmic rays, reject them, and replace them with clean data from a different exposure, will effectively remove a majority of the cosmic rays that impacted the detector. STScI’s data analysis and image processing software, known as astrodrizzle, will easily handle this kind of computation to create a clean image from multiple exposures. The few cosmic rays that are leftover can be handled in photo-editing software with the offending pixels replaced by the average background level.  

Photobombs by asteroids or other satellites

A much less frequent impact on images is the occasional photobomb by distant asteroids, or even other satellites that happen to cross Hubble’s field of view during an exposure. Once again, these are transient events, meaning they will only affect a single exposure, and are easily calibrated out, but it is interesting in the case of asteroids to consider the distances involved in the image. While the asteroids certainly are distant objects, millions of miles away from Hubble, the subject of such images is usually a distant galaxy or galaxy cluster, many, many orders of magnitude more distant than the asteroid (millions to billions of light-years, and remember a single light-year is six trillion miles!). Similar to cosmic rays, these artifacts are usually calibrated out of the data during the pre-processing stage. If anything makes it to the final image, it can again be cleaned up by replacing those pixels with the average local background level.

It’s a dirty job, but somebody’s gotta do it

This list of artifacts is certainly not comprehensive, but in my experience, represents the vast majority of issues an image processor can expect to face when pulling Hubble data together into a composite image. Hopefully, along with slaying ghosts and dragons, I’ve also slain any doubts you may have about Hubble images. When the James Webb Space Telescope launches later this year, we know we’ll see similar artifacts to what we’ve encountered with Hubble just based on the imaging technology. However, it will be interesting to see how the Webb point-spread-function and diffraction spikes will affect bright stars. Webb’s mirrors and optical assembly are very different from Hubble’s and will certainly impart its own unique signature on the imagery. As we’ve demonstrated many times in this blog, the process of creating astronomical imagery is a careful balance of art and science. Dealing with these artifacts is itself a balance of art and science, and the removal of these artifacts does not imply that the images are fake in any way. Creating an image free of distractions, which maintains the integrity of the source data and helps to illustrate the science results of a press release, is the ultimate goal of the image processor. Much like a documentary photographer, my job is to provide a straightforward and accurate representation of the data. There are times, like in the case of Eta Carina, where even the diffraction spikes of a bright star may prove to be an unnecessary distraction from the science and will require removal or mitigation. It is important to remember that even though our eyes alone cannot see what Hubble sees, that does not make these images any less real. Hubble gives us a window into the unseen universe and in doing so allows us to gain a deeper understanding of the cosmos and our place within it.

Here is a list of links to the press releases for all photos used as examples in this post:
1. UGC 2885 (charge bleed, ghosts, diffraction spikes)
2. Supernova Remnant E0102 (CTE, cosmic rays)
3. Serpens Nebula HBC 672 – Bat Shadow (bad pixels, death star)
4. Eagle Nebula (detector field of view)
5. NGC 2020 (mosaic artifacts)
6. Lagoon Nebula (dragon’s breath, ghosts)
7. Photobombing Asteroids (self-explanatory)