Home / Learn / Science / Sky watching / Lesson 6.7
All lessons

Module 6: Astrophotography · Lesson 6.7 · about 25 min

Processing honestly: stretch, colour and what is real

Astronomical data are linear and mostly dark; a stretch makes the faint parts visible. Colour comes from filters and is often mapped from invisible wavelengths. Honest processing changes how data look, never what they say.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 6 · about 25 minutes · read first: 6.4 Stacking: why many short shots beat one long one, 6.3 Calibration frames: darks, flats and bias · maths: functions and curves. You’ll be able to explain linear data and stretching, describe how colour images are made, and tell honest processing from fabrication.

Astronomical data are linear and mostly dark, so a stretch is needed to show the faint parts; colour comes from filters and is often mapped from invisible wavelengths, and honest processing changes how data look, never what they say.

1. Why this matters

Astronomical images shape how the public imagines the universe, and professional images are data first. Knowing what each processing step does lets you make beautiful images that are also true, and read other people’s images critically.

2. Linear data and stretching

A camera records light linearly: twice the photons, twice the number. A galaxy’s core may be 1,000 times brighter than its outer arms. Shown linearly, with the core set to white, the arms land near black.

A stretch is a curve that maps recorded values to display values, lifting faint values much more than bright ones. Common curves: a gamma curve (like a phone camera), and the arcsinh stretch (linear for faint values, logarithmic for bright ones, which keeps star colours). The black point sets which value becomes black.

Try it: stretch the data
The camera records light linearly: twice the photons, twice the number. Faint outer arms may hold under 1 % of the core's signal, so a linear display shows only the core. A stretch is a curve applied to every pixel the same way; it changes how the data look, not what was recorded. Honest processing applies the same curve everywhere, says what was done, and never paints in detail that is not in the data. Made-up galaxy.

Things to notice:

  • Linear: only the core shows.
  • Arcsinh with strength 40: the arms appear, the core is not burnt out.
  • Raising the black point too far clips faint data to black: information lost.

Process a stacked galaxy image:

Linear data. The stack looks almost black: the galaxy's faint arms are just 1% of the brightest star.
Set the black point. Subtract the sky level so the background sits just above zero.
Stretch. An arcsinh curve with β = 40 lifts 0.01 to 0.09, nine times brighter, while bright cores stay below white.
Colour. Combine red, green and blue filter images, or map invisible hydrogen, oxygen and sulphur light to colours (the "Hubble palette") and say so.
Honesty rules. Never paint in detail, clone out real features or add stars. Say what was done.
Maths and principle

Stretches (input x and output y from 0 to 1, after subtracting the black point b: x′=(x−b)/(1−b)):

ygamma=x′1/γ,yasinh=asinh⁡(βx′)asinh⁡β

Why arcsinh: asinh⁡(z)≈z for small z and ≈ln⁡(2z) for large z, so faint signal is lifted strongly while bright parts are compressed like magnitudes.

Colour-preserving stretch: apply the same factor to R, G and B of each pixel (computed from its total brightness), so a star’s colour ratio is kept.

Try it: with β=40, a value of 0.01 becomes asinh⁡(0.4)/asinh⁡(40)=0.390/4.382=0.089: nine times brighter on screen.

3. Where colour comes from

  • Broadband RGB: red, green and blue filters (or a colour sensor) give roughly natural colour after colour calibration, which can use stars of known colour.
  • Narrowband: filters that pass single emission lines: hydrogen-alpha (656 nm, deep red), oxygen III (501 nm, teal), sulphur II (672 nm, red). Mapping three of them to red, green and blue (the “Hubble palette”, SHO) is a false-colour image: it shows chemistry, not what the eye would see. Say so in the caption.
  • Invisible light: infrared, ultraviolet, X-ray and radio images (Astronomy 1.6) are always mapped to visible colours by choice.
Linear
Pixel values proportional to the light collected.
y = asinh(βx) / asinh(β)
Arcsinh stretch: lifts faint signal, keeps bright parts below white.
False colour
Colours assigned to wavelengths, possibly invisible ones.
Physical meaning
The camera records light like a ruler, but our eyes judge brightness like magnitudes, by ratios. A stretch translates the ruler into eye language so faint and bright parts both show. It changes the display, not the measurement, as long as you treat every pixel by the same rule.

“Colourful space images are fake.” Most use real data with honest choices: stretching to show faint parts and mapping real wavelengths, some invisible, to colours. Good images state their filters and processing. Fake would mean adding things that are not in the data.

4. Honest processing

Fair: calibration, stacking with outlier rejection, a global stretch, colour calibration, gentle noise reduction and sharpening applied evenly, removing gradients from light pollution, and stating what you did.

Not fair: painting or cloning in detail, pasting stars or objects from other images, AI “enhancement” that invents structure, or presenting false colour as natural. If in doubt, describe your steps in the caption, as professional image releases do.

5. Worked example

How much does a stretch lift faint arms? Core at 0.9, arms at 0.004 above a black point of 0.002.

  • Linear display: arms (0.004−0.002)/0.998=0.002: 0.2% grey, invisible.
  • Arcsinh β=100: asinh⁡(0.2)/asinh⁡(100)=0.199/5.298=0.0375; with gamma 2.2 on top for display, about 0.22: clearly visible. The core maps to about 0.99: still bright, not clipped.

6. Practice

More practice (answers below):

  1. Is an image with hydrogen-alpha shown in green “natural colour”?
  2. What happens to faint data if the black point is set above their level?
  3. Why should the same stretch apply to the whole image?
Answers
  1. No: it is false colour (hydrogen-alpha is deep red).
  2. They become pure black: the information is clipped away.
  3. A local stretch can create or hide structure; a global curve keeps the relative brightness order of every pixel.
  • Raw astronomical data are linear and need stretching to view.
  • Arcsinh stretches lift faint signal while protecting bright parts.
  • Colours come from filters; narrowband palettes map invisible light.
  • Honest processing never adds, removes or invents features, and is documented.

Try it with your own files: open FITS images and try linear, log and arcsinh stretches in the FITS viewer.

New words: linear data, stretch, gamma, arcsinh stretch, black point, colour calibration, narrowband, false colour, Hubble palette.

Check yourself

1Why does an unstretched (linear) astronomical image look almost black?
2In the Hubble palette, which colour usually shows oxygen (O III) light?
3Which step is NOT honest processing?

Sources

  • Howell, S. B. (2006), Handbook of CCD Astronomy, 2nd ed., Cambridge University Press (the CCD equation, calibration frames)
  • Berry, R. and Burnell, J. (2005), The Handbook of Astronomical Image Processing, 2nd ed., Willmann-Bell
  • Lupton, R. et al. (2004), Preparing red-green-blue images from CCD data, Publications of the Astronomical Society of the Pacific 116, 133 (arcsinh stretch)

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Numbers in the examples are made up to show the idea unless a source is given. Simulations run in your browser on made-up data; nothing you type is sent anywhere. Astronomy only: nothing here is astrology.