# Do you know about any Image Processing Challenges known in Space Telescope like Hubble?

For a computer science project in University;

I need to find image processing problems on space exploration and Space Telescopes, only related to the image itself.

I have wrote a computer program that is able to separate every pixel of an image according to its pixel intensity (the whitest to the darkest) in 3 (or even more) big groups.

I was thinking that this might be somehow useful in a space image processing situation in a well known challenge of the technology in the Telescopes or machine vision devices around space.

Situation can be very specific, just as an example calculating the mass of a star by the look of the pixels which compose it, or something like this.

• Are you asking for problems related to the real problems in modern research, or just more general processing things that you could do with an image? (At the moment it's not clear if you need a "Hubble type" pro application or you're just looking for any interesting image processing exercise to do.) – Andy Nov 16 '16 at 9:55
• Please specify if you're asking about data processing challenges onboard of a satellite, due to limited downlink, or something else. – AtmosphericPrisonEscape Nov 16 '16 at 11:03
• Many thanks, @Andy yes, I'm looking for a real problem to serve as material for a research, if you can mention them this will help me to orientate my searching on internet. If it's a real issue present not only in the Hubble but in other "astronomic" and space exploration imaging, it'd be great. – sujeto1 Nov 16 '16 at 11:40
• @AtmosphericPrisonEscape you quite aimed what I'm intending to do, in fact it's a digital design (a chip?) that can achieve a function on computer vision or image processing Onboard the satellite. My design perform "image segmentation" of a picture, processing pixel by pixel. – sujeto1 Nov 16 '16 at 11:40
• You might want to look at ImageJ , imagej.nih.gov/ij ,for general feature extraction, and the many NASA tools available at software.nasa.gov/data_and_image_processing – Carl Witthoft Nov 16 '16 at 13:45

First, I'll talk a little bit about the state of what is being done: (according to my knowledge)

In Astronomy on-board processing is usually avoided if possible.

The reason for this is to have the pure data as unspoiled as possible, because instrument effects, cosmic rays, the light from Earth's corona, etc. can introduce many systematic errors into the data. If that would be removed on-board and then sent to Earth, then there would be no control later on what the actual, original data was.
This became relevant e.e. for the exoplanet-finding Kepler(see also the technical data there) space telescope where the photometric data was showing very unexpected behaviour at first, before the reasons for it were realized, obviously long after the spacecraft was designed and built.

Thus, the image processing challenges are usually resolved on the ground stations and with the astronomers on their big computers. Also the interpretation of the data, after being cleaned, debiased, etc. is non-trivial.

As however the downlink speed depends on the distance $d$ to Earth like $\sim \frac{1}{d^3}$ so satellites that orbit far away do have to apply some clever ways to reduce data sometimes. The Hubble is not a candidate for this, as it orbits Earth at only some 500 km a.s.l, its data can be transferred to Earth without problems.

The Gaia satellite however, orbits the second Earth-Sun Lagrange point and is thus farther away from Earth, 1.5 million km. This mission is an astrometric mission, meaning it's there to measure positions and velocities of stars to extreme precision. If all the pixels of all onboard systems would continuously be read out and produce data, that would be too much to handle and some would be lost. So Gaia has a pre-detector that tells the main detector "Hey a star is coming". The main CCD then creates a moving window of 12x12 pixels from which all electrons are read-out and accepted as data. So that's some limited processing there.

In terms of data transfer, missions like New Horizons (Pluto encounter last year) or the Voyagers, are too far out to simply send their data over. Therefore the Deep Space Network exists, to look out for radio signals from those probes and actively amplify them.

What you could do:
So now that I've spent time explaining that on real spacecraft it's not being done what you want to do, we can bend the rules and just try to think about what's being done on the ground (if you want to):

• Image calibration: A detector taking an image wants to be very well understood, before claiming any interesting discoveries or measurments. Part of this process is taking a dark image to characterize what your CCD is doing when nothing is measured. Then a flatfield image is done, an image of an ideally uniformly white surface, in order to characterize pixel-to-pixel variations. More. For the measured image then, the dark image is substracted and divided through the flatfield image.
• Cosmic ray removal: Cosmic rays fly through your camera in a random direction at random times (in space as well as on the ground). They produce streaks in CCD cameras, and can be automatically removed by taking the same image three times, and discarding one of them or only a pixel-region if a dramatic intensity streak somewhere is measured.
• Measuring a point-spread function: Stars are so far away, that they appear as point sources. However through instrumental effects this ideal point will always be spread out into a finite shape, called the PSF. Because this function can be easily a few pixels wide, it is very important to know the shape of the PSF to be able to assign photons to a star that have been measured in off-center pixels. Knowing then a star's photon output and it's distance one can calculate the luminosity. But even that can't be done on an on-board chip. What you could to is to try to automatically fit an Airy-function to the PSF.

Ok so that was just a few ideas, but I'm not an observer, there may be better ones out there. Also you got keywords not to google further.

Based on the original question...

which I feel like is a little late now, as most universities are finished until next year.


isn't as involved or difficult as it has been presented.

If all you're looking for are image processing problems, then you're going to need to look into the following:

1. Bias levels
2. Dark levels
3. Background levels
4. Cosmic Rays

Anything else regarding the images ie: stacking of filter combinations, are secondary and rarely required for any science case.

The funny thing is, these are common for any CCD image with a long exposure, so the space telescopes bit only adds more chance for cosmic ray interactions.

Enjoy