This question may seem vague, but I'm far from an expert at photography/photometry/etc, so please bear with me...

But, let's say for instance I wanted to compare two images of the sun from differing telescopes-- inherently causing the image sensors/ccd/software to likely be different. Thus comparison of images takes place at the RGB 1024x1024x3 3-dimensional matrix level through MATLAB, a matrix-based language.

Since, for instance, I can retrieve digital images from different observatories that represent nearly the same image when analyzed qualitatively, where do the cmos/ccd/photon sensors cause differences at-- if any at all? As the sun is so large in size and the observatories are viewing relatively the same side/angle of the sun, what hardware/software differences would cause thus matrices to differ?

Here's what I've thought of so far; some inferences and some researched:

  1. The demosaicing algorithms may differ across units.
  2. Since all cameras have a sensor noise that is related to the type of camera it is and this can be used to identify the type of camera that took the image--as discussed on How can I tell if two pictures were taken with the same camera without metadata? --then differing sensors should cause differing noise values and thus differing RGB matrice values.
  3. The file format and data transfer from each telescope may differ, causing data loss, and maybe other issues. But a consistent raw format(or file format in general) may combat this, or so I assume.
  4. Since MATLAB converts these images to the aforementioned matrices, then do these conversions invoke other issues as this isn't the native means of image representation on a computer?


Much appreciation for the long, very detailed answer from @jrista

I'm sorry for the confusion, but I'm not hoping to identify the type of camera based upon their noise, metadata headers, or anything of the sort, but rather the relatability amongst differing cameras on different telescopes, hoping that there is, in fact, a standardization of equipment as there is for image file format.

Anyways to elucidate on what I'm hoping to do and why I'm wondering such a question, I will further explain what I'm hoping to find out and why a somewhat "standardized" camera--containing data values that are relatively similar--is something I require. I don't mean to be too detailed or require answers for my project; I just think the confusion will persist without an in-depth explanation as to why I want to know such things.

I will be taking in FITS data from many ground-based solar observatories as well as FITS data from many space-based solar observatories. In doing so, I hope to use image differencing techniques--subtracting the space-based solar observatories' image from the ground-based one--yielding some RGB matrice values. The images will, in fact, be from the same time frame and I will be using a computer algorithm to determine the displacement/misalignment of the images, as the alignment of the two images is imperative to the success of my research. I will also be experimenting with a normalization function I wrote and other averaging and collectively image processing techniques I find useful to ultimately find a consistency amongst these values regardless of the telescopes in which such data was collected. Astronomical seeing distortions cause many problems with ground-based observations and I hope to quantify such seeing distortions through their pixelation values yielded via the differencing. I understand astronomical seeing, have taken into account the many variables that can AND will come up, and so I don't need an explanation as to why this will fail unless you're absolutely sure due to some fact about cameras/sensors and how their images will inherently cause some difference that I'm ignorant too. I've created plans to combat such variables.

All in all, my question as concise as possible--

Will differing cameras' images(at the pixelation value) be inherently different regardless of if the image is the exact same object?

  • 1
    \$\begingroup\$ I'm voting to close this question as off-topic because this question is more about sensor issues in signal processing than photography. This question should be migrated to DSP. \$\endgroup\$
    – scottbb
    Dec 12, 2016 at 2:14
  • 1
    \$\begingroup\$ You're question isn't dumb, but it is vague. What sort of information are you trying to discern from different images of the same object. Also, are these images taken at or at least near the same time? The reason I ask is that image differencing can be somewhat problematic. The results will depend on what you're trying to understand about the images. You're 4 points are actually quite good and accurate! \$\endgroup\$ Dec 12, 2016 at 2:56
  • \$\begingroup\$ @scottbb I will be happy to post the same question on that forums, but for my best chances of having the question answered I believe that this forum will also be a good outlet. I will, however, transfer it over if it is merely a signal processing question! \$\endgroup\$
    – Maddux123
    Dec 12, 2016 at 18:25
  • \$\begingroup\$ Don't worry about moving it yourself; cross-posting is generally discouraged at SE. If it belongs at DSP, it will get migrated by a mod. If not, it will stay here. \$\endgroup\$
    – scottbb
    Dec 12, 2016 at 18:34
  • 1
    \$\begingroup\$ I am going to keep this here for the moment. I am going to try to answer...as I believe I can at least help in some cases. However it may still end up being moved to another SE site. \$\endgroup\$
    – jrista
    Dec 13, 2016 at 4:24

1 Answer 1


Round two!

If you are looking to compare the results of space-based images with ground-based images, then you have a whole lot of variables to deal with. Ignoring the atmospheric issue for the moment, the question really is not about pixels per-se. You are going to have differences due to pixel size, as well as scope focal length, scope aperture, overall field of view, etc.

Given you will be working with solar images, I am going to assume for the moment that the data will be well into the realm of shot noise limited. A solar imaging system, even one that uses a doppler tuned etalon to limit bandpass to the sub-angstrom level, should still be capable of producing strong signal, more than enough to render camera noise effectively moot. The only real exception might be if you are trying to measure solar granule motion in real time or something like that, which would require very short exposures. So I don't think that camera noise, or even camera FPN, is really going to be your largest concern.

The question has to do with image scale and FoV, really. Barring the unlikely scenario that all of your imaging systems are truly identical, with the only variable being whether they are Earth-bound or in space, image scale and FoV are going to be your primary variables that affect the data.

Image scale is relative to the pixel size and focal length. The actual formula is:

ImageScale = (206.265*PixelSize)/FocalLength

Where PixelSize is in microns, and FocalLength is in millimeters. If you have a 5 micron pixel and a 1000mm (1-meter long) focal length, then you have an image scale of 1.031"/px, or about one arcsecond per pixel.

The higher the image scale (which is actually a smaller image scale number), the higher the resolution of the system unless you have exceptionally bad seeing. A long telescope with small pixels is going to resolve considerably more than a short telescope with large pixels. Even if the two scopes use the same sensor with the same pixel size, the longer telescope is going to spread details over more pixels, so they will be more accurately resolved. That will usually result in discrepancies once the disparate data is registered.

The FoV is going to be dependent on the sensor size as well as the focal length. If you had 5 micron pixels, with a 1024x1024 pixel sensor, your FoV would be ~17 arcminutes, or about one quarter of a degree, on either side. A 500mm (half meter long) telescope would give you an FoV of about half a degree, a 2000mm (two meter long) telescope would give you an FoV of about 1/8th of a degree. Professional telescopes tend to have longer focal lengths, however I honestly don't know what kind of space-based systems you might be using or what focal lengths they may have.

Now it is possible to register images with disparate image scales and fields of view. The registration algorithm will identify the stars in each image relative to a single reference, and adjust each frame accordingly. Advanced registration algorithms can also correct for distortion within the image, as well as perform simple translations and rotation. The thing about registration, however, is it will change the nature of the data. Depending on exactly what aspects you correct, you may end up with interpolation artifacts, and worse, they can be non-uniformly applied throughout the field.

Registration issues can be worse if you choose a poor reference frame. Some registration tools allow you to plate solve your images and generate a synthetic starfield from the plate details. You can then use the synthetic, distortion free and accurately modeled starfield as your registration reference, which can minimize compounding artifacts due to poor reference selection. However, it will not eliminate them.

If your individual images are indeed shot noise limited, then I honestly do not see the camera noise really being an issue. Every camera has FPN and DFPN. The former could limit SNR if your exposures are too bright, and the latter could limit SNR if your exposures are too dim. However proper calibration with flats and darks should correct most FPN, leaving you with just the temporally random noise. Even if you had 8-10e- worth of read noise and a couple e- worth of dark current, if your signal is even just 200e- or greater, you would completely swamp the camera noise, rendering it effectively meaningless.

  • \$\begingroup\$ I think I clarified the question to the best of my ability. However, to answer your question asking if MATLAB is only capable of converting these values to matrices, etc. ... It has the functionality built into it to read the "primary data" of FITS files, converting them to double precision matrices after a scaling of the image by the slope and intercept values fitsread. Would you say that suffices as reading the original data as is or no? I can also use the 'raw' tag to prevent scaling, which I'll likely employ. Thanks again! \$\endgroup\$
    – Maddux123
    Dec 15, 2016 at 18:54
  • \$\begingroup\$ I just rewrote my answer to account for your true question. Let me know if that answers your question. If this is solar imagery, I would suspect the median signal should be quite high relative to other forms of astrophotography. That should render camera noise effectively moot. You'll have other issues to contend with, though. ;) \$\endgroup\$
    – jrista
    Dec 15, 2016 at 19:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.