What Is Image Integration?
If you have spent any time in astrophotography, you have heard the advice: “take lots of exposures and stack them.” But why does this work? What is actually happening when you combine dozens or hundreds of sub-frames into a single image? The answer lies in a concept called image integration, and understanding it will change how you plan your imaging sessions.
Image integration is the mathematical process of combining multiple photographs of the same target into one master image. When done correctly, it increases signal while reducing noise, revealing detail that is quite literally invisible in any single sub-frame.
Signal vs. Noise: The Fundamental Battle
Every photograph you take contains two things:
- Signal – The actual light from your target (galaxies, nebulae, stars) plus sky background light
- Noise – Random variations from your camera sensor, thermal effects, and the random nature of light itself (shot noise)
The critical metric in astrophotography is the signal-to-noise ratio (SNR). A higher SNR means a cleaner, more detailed image. A low SNR means a grainy, washed-out result. The entire goal of image integration is to maximize SNR.
Why Stacking Works: The Math Is Simple
Here is the key principle: signal adds up linearly, but noise adds up as the square root.
When you combine N exposures:
- Signal increases by a factor of N
- Noise increases by a factor of √N (the square root of N)
- SNR improves by a factor of N ÷ √N = √N
This means stacking 4 frames gives you 2x better SNR. Stacking 9 frames gives you 3x. Stacking 100 frames gives you 10x. The improvement follows a classic square-root curve.
| Number of Frames | SNR Improvement |
|---|---|
| 1 | 1x (baseline) |
| 4 | 2x |
| 9 | 3x |
| 16 | 4x |
| 25 | 5x |
| 100 | 10x |
| 400 | 20x |
Notice the pattern: to double your SNR, you need four times as many frames. This is why astrophotographers spend multiple nights on a single target. There is no shortcut past the math.
The Diminishing Returns Problem
That square-root relationship has a brutal consequence: diminishing returns. Going from 1 frame to 4 frames doubles your SNR. Going from 100 to 104 frames makes almost no visible difference.
This is why experienced imagers have a target integration time in mind before they start. For most deep-sky objects with a moderate setup, the sweet spot is somewhere between 3 and 10 hours of total integration, depending on the target’s brightness, your focal ratio, and your light pollution levels.
Bright targets like the Orion Nebula or Pleiades can look excellent with 1 to 2 hours. Faint targets like dark nebulae or distant galaxies may need 10 or more hours to reveal meaningful detail.
Read Noise: Why Many Short Exposures Can Beat Few Long Ones
Every time your camera sensor is read out, a small amount of random noise is added. This is called read noise, and it is a fixed cost per frame regardless of how long the exposure is.
Consider two approaches for 1 hour of total integration:
- 60 frames × 60 seconds each: 60 instances of read noise
- 4 frames × 900 seconds each: 4 instances of read noise
The longer exposures have less total read noise, which is better. But there is a catch: if your individual sub-frames are too long, you risk overexposing bright stars, dealing with satellite trails and aircraft, and accumulating more noise from light pollution and sensor heat.
The practical answer is to find your optimal sub-frame length – long enough that read noise becomes negligible compared to shot noise from the sky background, but short enough that you can manage quality control. For most backyard setups under typical skies, this is somewhere between 60 and 300 seconds per frame.
A simple rule of thumb: if your sky background is bright enough that each sub-frame is sky-limited (the sky background noise exceeds your read noise), then adding more frames of the same length always helps equally.
Stacking Methods: Mean, Median, and Sigma Clipping
When it comes time to actually combine your calibrated frames, your stacking software (Siril, DSS, PixInsight) offers several methods:
Average (Mean) Stack
Simply adds all frames and divides by the count. This gives the best theoretical SNR improvement. However, it is vulnerable to outliers – a single airplane trail or cosmic ray hit will contaminate the average.
Median Stack
Takes the middle value at each pixel across all frames. Very effective at rejecting outliers like satellite trails and hot pixels. The trade-off is that median combining has slightly lower SNR than averaging – you lose about 20 percent efficiency compared to mean stacking.
Sigma Clipping / Winsorising
The best of both worlds. These algorithms iteratively identify and reject outlier pixels, then average what remains. Sigma clipping is the default recommendation in Siril and PixInsight for most deep-sky work because it provides near-optimal SNR while automatically removing artifacts.
A Real-World Example: 6 Hours on M33
Consider this image of the Triangulum Galaxy (M33), captured with 120 individual 180-second exposures across three nights. That is 6 hours of total integration time with a Canon EOS Rebel T3i and an 8-inch Newtonian reflector.
The Triangulum Galaxy (M33). 120 × 180 second exposures stacked in Siril. The spiral arm detail visible here is simply not present in any single sub-frame.
With 120 frames stacked, the SNR improved by a factor of about 11 compared to a single frame (√120 ≈ 11). That is the difference between seeing a faint gray smudge and resolving actual spiral arms, H-II regions, and dust lanes.
Or Take the Heart Nebula: Five Nights, Two Panels
This next image represents an even more ambitious integration. The Heart Nebula mosaic required 85 exposures of 180 seconds each, captured across five separate nights, stitched together as a two-panel mosaic in Siril.
The Heart Nebula (NGC 896). A two-panel mosaic totaling 85 × 180 second exposures over five nights.
Each panel was stacked separately (roughly 42 frames per panel), giving an SNR improvement of about 6.5x per panel. The mosaic stitching then combined both panels into a wider field of view. This is image integration applied twice: first to build each panel, then to merge them into a seamless whole.
Practical Takeaways for Your Imaging
- More frames is almost always better – but plan around the square-root curve. Going from 10 to 20 frames is a 41 percent SNR boost. Going from 100 to 110 frames is less than 5 percent.
- Find your optimal sub-frame length – long enough to swamp read noise, short enough to manage quality. Test with your own gear and skies.
- Use sigma clipping for stacking whenever possible. It gives you average-level SNR with median-level outlier rejection.
- Calibration matters – darks, flats, and bias frames remove fixed-pattern noise so that what remains is truly random (and therefore reducible by stacking). See our complete guide to calibration frames.
- Total integration time is what counts – whether you get there through many short exposures or fewer long ones, the final SNR depends primarily on total integration time (assuming you are sky-limited).
The Bottom Line
Image integration is not magic. It is statistics. But it is the single most powerful technique in all of astrophotography. Understanding the square-root law of noise tells you exactly why that faint galaxy needs another three hours of data, and why your tenth frame looks dramatically better than your first while your hundred-and-tenth looks almost identical to your hundredth.
The next clear night, point your camera at something faint, collect as many frames as you can, and watch the noise melt away as the signal emerges. That is the magic of mathematics applied to starlight.
Related reading:
