How images are made
Are Hubble Images Real? Image Artifacts and How They Are Cleaned
Spikes, ghosts and streaks in raw Hubble data come from the telescope, not the sky. Knowing where each comes from is how they get removed honestly.
Ghosts, spikes and dragon's breath sound like a fantasy series. They are names for marks that the Hubble Space Telescope itself leaves in its raw data, and if they stayed in a published picture they could be mistaken for real structures in space. Removing them is part of making a press image, which raises a fair question: if an image has been cleaned, is it still real? The answer starts with knowing what each artifact is and where it comes from. The companion article, Are Hubble Images Real? What Makes Them Real, takes the question from the other side: what the colors and shapes in a finished image actually record.
The long path from sky to detector
Light entering Hubble travels down the main baffle, a long tube, to the 8-foot (2.4-meter) primary mirror, which reflects it back up to a smaller secondary mirror held in the middle of the tube by cross-shaped struts. The secondary sends the light back down a central baffle to the instruments mounted behind the primary, where small mirror assemblies belonging to each science instrument pick off the part of the field they need. Inside an instrument such as Wide Field Camera 3, more mirrors guide the light through a filter and onto the detector.
Every surface along that route is a chance for stray light to bounce where it should not, and every part of the detector has its own habits. The artifacts below fall into three groups: those made by the detector, those made by the optics, and those that arrive from outside.
Artifacts made by the detector
Charge bleed. Hubble's visible-light cameras use charge-coupled devices, the same basic technology as a phone camera. A CCD pixel works like a bucket catching raindrops: it collects photons during the exposure, and at the end the contents are passed from bucket to bucket, a bucket brigade, to be read out. A very bright star fills its buckets to overflowing, and the excess spills along the column into neighboring pixels, leaving a bright streak. The preferred fix is to replace the streak with real data from another exposure taken at a different rotation. Failing that, a clean part of the star's glow at right angles to the bleed can be copied, rotated 90 degrees and blended in, without adding anything that was not recorded.
Charge transfer efficiency. As a CCD ages in space, some pixels stop passing on all of their charge at once and dribble it out during readout, so stars grow faint tails pointing back along the readout direction. The correction is built into the standard calibration pipeline, which pushes the stray charge back where it belongs. Anyone downloading Hubble data from the MAST archive will rarely see the effect.
Chip gaps. The Wide Field Channel of the Advanced Camera for Surveys and the UVIS channel of Wide Field Camera 3 each join two CCDs, with a narrow strip of no data between them. Observers usually shift the telescope slightly between exposures so the gap falls on a different part of the sky each time. Where that was not done, the gap can be filled from other Hubble data or even ground-based data; if nothing real exists, the image is cropped, or as a last resort the strip is set to the average background level.
Bad pixels. Some pixels are hot, always reporting a high signal; others are dead and report nothing. They are mapped and handled in calibration. The infrared channel of Wide Field Camera 3 has a well-known circular patch of dead pixels, nicknamed the death star, that is marked in the planning software so observers keep important targets off it. Combining several slightly offset exposures fills such holes with good data.
The stair-step. Detectors can shape the whole frame. The Wide Field and Planetary Camera 2, retired in 2009, had four CCDs in a square, one of them the Planetary Camera with pixels covering half the angle on the sky. Combined into one picture, that chip shrinks relative to the others and leaves a stepped notch in one corner.
Many WFPC2-era images, including the 1995 Pillars of Creation described on the Eagle Nebula page, kept the notch.
Mosaic seams. Because a single frame covers a small piece of sky, big targets are tiled. The 30th anniversary image of NGC 2014 and NGC 2020 needed a grid of four by three pointings in each of four filters, 48 images in all. Even after careful calibration, the overlaps between tiles need local brightness and contrast matching, and small holes in the coverage are filled from other data or with the local background level.
Artifacts made by the optics
Diffraction spikes. The four spikes on bright stars in Hubble pictures are made by light bending around the struts that hold the secondary mirror. A camera lens stopped down to a small aperture does something similar with its iris blades. Different wavelengths spread by different amounts, which gives the spikes of color images their faint rainbow edges. Spikes are so tied to the idea of a star in a space photo that they are normally left in; removing them from a crowded field would also be close to impossible.
There are exceptions. In the 2019 press image of Eta Carinae, rays of starlight streaming through gaps in the dust around the twin lobes were a central part of the result, and some of the strongest rays happened to line up with the diffraction spikes of the bright central star. The spikes were removed so that nobody would read them as rays, or the rays as spikes.
Ghosts, donuts and glints. A bright star in or near the field produces reflections inside the optics. Near the center of the detector the reflection is a broad glowing ghost close to the star; farther off-axis it becomes a ring, a donut or a figure eight, displaced and distorted. Stars just outside the field can still send light down the tube and onto the detector as glints. One large, flame-shaped glint has earned the name dragon's breath. These are reduced with data from other exposures, by toning down the reflection, or, if all else fails, by setting the affected pixels to the background level.
Point-spread function. Hubble does not resolve the surface of distant stars; they are points. A bright point still spreads into a halo whose shape depends on the camera and filter, and in color images bright stars wear colored rings. The halo is part of the data and stays in, though it can be tamed with local adjustments if it swamps the surroundings.
Artifacts that come from outside
Cosmic rays. Above the atmosphere, Hubble is constantly hit by high-energy particles. Those that reach a detector leave a few saturated pixels or a streak, depending on the angle, and a single exposure can look as if it is snowing. Because each hit appears in only one exposure, software such as AstroDrizzle compares several exposures of the same field, rejects the hits and fills in clean data. The few that survive are fixed by hand.
Asteroids and satellites. Now and then something crosses the field during an exposure. Asteroids are the most common culprits in deep fields, millions of kilometers away against galaxies billions of light-years beyond.
In the Frontier Fields image of Abell 370, five previously unknown asteroids left curved and S-shaped trails. They bend because Hubble itself moves along its orbit during the exposures, a parallax effect, and because Earth and the asteroids are also moving. Trails like these are usually removed in pre-processing; this release kept them on purpose, and more examples are collected in Asteroids Photobomb the Universe.
The common thread is the order of repair. Real data from another exposure come first, then data from another telescope, and only then neutral background. Nothing is painted in that the telescope did not see. Choosing colors for the cleaned filter images is a separate step with its own rules, explained in Are the Colors Real?.
Questions readers ask
Are Hubble images edited?
They are processed. Exposures are calibrated, combined and colored, and instrument artifacts such as cosmic rays, chip gaps and reflections are removed, preferably by substituting real data from other exposures. Structures in the object are never added or invented.
What causes the spikes on stars in Hubble images?
Diffraction of light around the four struts that hold Hubble's secondary mirror. The pattern is a property of the telescope, not of the stars. It is usually left in, but was removed from the 2019 Eta Carinae image because it overlapped real rays of light.
Why do some old Hubble images have a stepped corner?
They were taken with the Wide Field and Planetary Camera 2, which had four detectors, one of them at twice the magnification. When the four frames were combined, that one appeared smaller and left a stair-step notch, as in the 1995 Pillars of Creation.
What is dragon's breath in a Hubble image?
A large, flame-shaped glow caused by a very bright star just outside the camera's field. Its light reflects inside the optics and lands on the detector even though the star itself is not in the frame. Image processors reduce or remove it.