A modern photograph is often not a single exposure. Phones and many cameras capture several frames, align and combine them, recognise faces, skies and objects, and adjust each area separately, all within a fraction of a second. This is computational photography: using computing to produce images that a small sensor or single exposure could not achieve alone. You met its everyday features in the mobile photography lesson; this lesson explains how the techniques work, where they appear in dedicated cameras too, and what their limits are.
What is computational photography?
Computational photography uses algorithms, and increasingly machine learning, to capture, combine and process image data in ways beyond a traditional single exposure. Everyday phone features such as night mode, portrait mode and multi-frame HDR were introduced in mobile and computational photography. Here we look at the underlying techniques.
Core techniques
| Technique | How it works | Result |
|---|---|---|
| Multi-frame stacking | Capture a burst, align frames, merge them | Less noise and more detail, like averaging in low-light exposure strategy |
| Multi-frame HDR | Combine frames with different exposures or many short frames | Detail in highlights and shadows, as in high dynamic range capture |
| Super-resolution | Use tiny shifts between frames, from hand movement or the sensor, to recover extra detail | Higher resolution than one frame |
| Depth mapping | Estimate distance using two cameras, special pixels, sensors or AI | Simulated background blur and relighting |
| Scene segmentation | Identify sky, skin, hair, food, text and other regions | Different processing for each area |
| AI denoising and detail reconstruction | Neural networks remove noise and rebuild fine detail | Cleaner images, sometimes with invented texture |
Computational features in dedicated cameras
- Pixel-shift high resolution: the sensor moves slightly between several exposures, then combines them for greater detail and color accuracy, ideal for still subjects on a tripod.
- In-camera focus stacking or bracketing, as in focus stacking.
- In-camera HDR and multiple exposure modes.
- Simulated ND or live composite modes that combine many frames to mimic long exposures, as in long-exposure planning.
- AI-based autofocus and scene recognition, as in modern camera technology.
The processing pipeline
- Capture: often many frames, sometimes before you press the shutter.
- Align and merge to reduce noise and blur.
- Analyse the scene: faces, skies, text, motion.
- Apply local processing: tone mapping, color, sharpening and noise reduction per region.
- Output: a finished JPEG or HEIF image, or a computational RAW file with some processing already applied.
Remember:Some computational RAW files are already merged from several frames. They offer more editing flexibility than a JPEG, but are not identical to a single-exposure RAW from a traditional camera, as compared in RAW versus JPEG and other image formats.
Strengths and limits
| Strengths | Limits |
|---|---|
| Cleaner low-light images from small sensors | Moving subjects can create ghosts or smearing |
| Balanced exposures in high-contrast scenes | A uniform "processed" look across many images |
| Convenient background blur | Errors around hair, glasses and complex edges |
| Sharp, detailed results without a tripod | Invented or over-smoothed detail at 100% |
Warning:Because computational processing can change or invent detail, consider whether it is suitable for scientific, evidence or strict documentary work, as in truth, manipulation and image integrity.
Common mistakes
- Assuming every photo is a single moment: many are merged from several.
- Using pixel shift with moving subjects: artifacts.
- Trusting AI detail completely: invented texture.
- Ignoring style options and RAW modes: accepting a look you did not choose.
How professionals use computational tools
Professionals use computational features where they help: pixel shift for product and art reproduction, in-camera stacking for macro, night modes for quick scouting and social content. They understand what the processing does, choose RAW or less-processed modes when control matters, check results at full size and avoid computational processing where strict accuracy is required.
Practical examples
A painting reproduction
- Situation
- A gallery needs a very detailed image of a painting.
- What to do
- You use a camera's pixel-shift mode on a tripod with steady lighting.
- Why it works
- Combining shifted exposures increases detail and color accuracy for still subjects.
- Result
- A highly detailed, accurate reproduction.
Ghosts in a night-mode photo
- Situation
- A night-mode image of a busy street shows doubled people.
- What to do
- You switch to a single exposure at a higher ISO for that scene.
- Why it works
- Multi-frame merging struggles with fast-moving subjects.
- Result
- A clean image without ghosting.
Key points
- Computational photography combines multiple captures and computing to exceed single exposures.
- Core techniques: stacking, HDR, super-resolution, depth maps, segmentation and AI denoising.
- Dedicated cameras offer pixel shift, in-camera stacking, HDR and simulated long exposures.
- Pipelines capture, align, analyse, process locally and output, sometimes as computational RAW.
- Watch for ghosts, edge errors and invented detail; avoid it where strict accuracy matters.
Frequently asked questions
What is computational photography?
What is pixel shift?
Is a phone photo a single exposure?
Can computational processing add detail that was not there?
Conclusion
Computational photography merges capture and computing to produce cleaner, sharper and more balanced images, with limits around motion, edges and invented detail. Next, you will learn how smartphone multi-camera systems work.
Finished reading? Track your progress through AI and new technology.