Chapter 23 — Quiz
Twenty-five self-check questions on computational photography: how phones stack frames, what HDR and tone mapping actually do, why night mode beats the dark, how portrait mode fakes shallow depth of field, the "phone look" pipeline, and — most importantly — when to trust the computation and when to override it. Mix of multiple choice, true/false-with-justification, short answer, and "what's wrong with this?" seeing-questions. Answers and one-line explanations are at the end. No peeking until you've committed to an answer — a guess you've reasoned through teaches more than a fact you've reread.
Multiple choice
1. The single foundational idea behind night mode, HDR, and multi-frame noise reduction is: - (a) a larger sensor that gathers more light - (b) a faster lens with a wider aperture - (c) capturing multiple frames and intelligently combining them into one - (d) a longer single exposure
2. Stacking many frames of the same still scene reduces noise because: - (a) the camera throws away the dark frames - (b) the real signal reinforces across frames while the random noise averages toward zero - (c) each frame is sharpened before merging - (d) noise only appears in the first frame
3. HDR's merge step solves which problem? - (a) a scene whose brightness range exceeds what one exposure can hold - (b) a subject that is moving too fast to freeze - (c) a background that won't blur on a phone - (d) a white balance that is too warm
4. Tone mapping is the step that: - (a) captures the multiple exposures - (b) compresses a wide captured brightness range down to fit a display or print - (c) identifies faces and skies in the image - (d) aligns the frames before merging
5. The flat, grey, over-detailed, haloed "HDR look" is caused by: - (a) too few frames in the merge - (b) heavy-handed (especially local) tone mapping - (c) shooting RAW - (d) a sensor that is too large
6. Night mode captures many short frames rather than one long exposure primarily to: - (a) save battery - (b) freeze your hand-shake (so each frame is sharp) and protect bright highlights from piling up - (c) make the file smaller - (d) avoid using the flash
7. Night mode's single greatest weakness is: - (a) bright daylight - (b) cold weather - (c) anything in the scene that moves during the capture - (d) vertical orientation
8. A depth map records, at each pixel: - (a) the color of the scene - (b) the brightness of the scene - (c) the distance of that part of the scene from the camera - (d) the white balance
9. Synthetic bokeh is: - (a) a real optical blur from a wide aperture - (b) a digitally simulated background blur applied according to the depth map - (c) a type of noise reduction - (d) the warm color phones add to skin
10. Portrait mode exists fundamentally because: - (a) phone software is lazy - (b) a phone's tiny sensor and small lens physically cannot blur a background much optically - (c) RAW files are too large - (d) people dislike sharp photos
11. The artifact most associated with portrait mode failing is: - (a) a green color cast - (b) a "hair halo" or snapped cutout edge where the depth map mis-segments the subject - (c) blown-out highlights - (d) light trails
12. Semantic segmentation in the phone pipeline means: - (a) cropping the image into a grid - (b) the phone identifying what regions are (sky, face, foliage, food) and processing each differently - (c) compressing the file - (d) merging exposures for HDR
13. The "watercolor" smear — waxy skin, mushy foliage — is the fingerprint of: - (a) over-sharpening - (b) aggressive noise reduction smoothing fine texture along with the noise - (c) HDR tone mapping - (d) a depth-map error
14. The single most powerful way to take control back from the entire computational pipeline is to: - (a) increase the screen brightness - (b) shoot RAW (the relatively unprocessed sensor data, before the pipeline's choices) - (c) use the front camera - (d) turn off the grid lines
15. For a moody, deliberately dark low-key shot (a silhouette, a single light), you should: - (a) turn HDR on to open every shadow - (b) turn HDR off (or shoot one dark exposure) so the shadows stay dark and the mood survives - (c) always use portrait mode - (d) use night mode
True / False — and justify your answer in one sentence
16. The JPEG your phone produces is an accurate record of what the sensor actually saw. True / False — and why?
17. A bigger background-to-subject distance makes portrait mode's synthetic bokeh more convincing. True / False — and why?
18. Maxing out the simulated aperture (the blurriest setting) always makes a portrait-mode shot look more professional. True / False — and why?
19. In a backlit scene, you usually get the most natural HDR by exposing for the bright highlights and letting the merge lift the shadows. True / False — and why?
20. Night mode will give a tack-sharp result of a person dancing at a dim party, just like it does for an empty street. True / False — and why?
Short answer
21. In two or three sentences, explain the difference between HDR's merge step and its tone-map step, and which one is responsible for whether the final photo looks real or cooked.
22. Name three distinct sources a phone can use to estimate the depth map for portrait mode, and state in one phrase what each is.
23. List, in order, four stages of the "phone look" pipeline that the phone applies to nearly every frame, and name what each one does to the image.
What's wrong with this? (seeing-questions)
24. A reader writes: "I braced my phone on a wall and shot a beautiful empty lit alley in night mode — it came out clean and bright. But when I shot my friend standing in that same alley, also in night mode, her face is a soft, half-transparent blur. My phone must be defective in low light." What is actually happening, and what are two ways to fix the friend's blur without changing phones?
25. A reader writes: "I'm photographing a hand-thrown ceramic mug to sell online. On my phone the glaze looks a vivid teal and the photo is super sharp and bright — but when the buyer received it they complained the real mug is a muted grey-green and the surface has subtle tool marks that don't show in my photo." Name the two computational pipeline behaviors causing the mismatch, and prescribe the fix.
Answer Key
1. (c) — Multiple frames intelligently combined is the foundational move; everything else in the chapter is a variation on it. (§23.1)
2. (b) — Signal is consistent across frames and reinforces; noise is random and averages toward zero. (§23.1, Figure 23.1)
3. (a) — The merge holds a scene's full dynamic range (e.g. shadowed face and bright window). (§23.2)
4. (b) — Tone mapping compresses the wide captured range to fit the narrower range a display can show. (§23.2)
5. (b) — Heavy local tone mapping crushes tones toward the middle, jacks local contrast, and leaves edge halos. (§23.2, 🔬 The Physics)
6. (b) — Short frames freeze hand-shake (each is sharp) and let the algorithm hold highlights instead of piling light onto them; alignment then combines them. (§23.3)
7. (c) — Motion breaks the core assumption that every frame is the same scene, so moving things ghost or smear. (§23.3)
8. (c) — A depth map records distance per pixel, not color or brightness. (§23.4)
9. (b) — Synthetic bokeh is a digitally simulated blur applied by the depth map, imitating a fast lens. (§23.4)
10. (b) — It substitutes math for a genuine optical limit: a small sensor/lens can't blur a background much. (§23.4, 💡 Why It Works)
11. (b) — Hair halos and snapped cutouts come from the depth map mis-segmenting the subject's edge. (§23.4)
12. (b) — Segmentation identifies what regions are and applies a tuned recipe to each. (§23.5)
13. (b) — Noise reduction can't always tell noise from fine texture, so it smooths both → the "watercolor" look. (§23.5)
14. (b) — RAW is the sensor data before the pipeline, returning full control. (§23.6)
15. (b) — HDR would open the very shadows that create the mood; turn it off (or shoot one dark exposure) to protect them. (§23.2, §23.6, Figure 23.7)
16. False — The JPEG is a heavily processed interpretation (color boost, tone map, sharpening, noise reduction, segmentation tweaks), not a faithful record; shoot RAW for that. (§23.5, §23.6)
17. True — More real depth difference gives the depth map an easier, cleaner edge to read, so the synthetic blur becomes believable. (§23.4)
18. False — Maximal blur exaggerates every segmentation error (halos, snapped cutouts, wrongly-blurred hands); a moderate blur is usually more convincing. (§23.4)
19. True — Blown highlights have no data to recover, while dark shadows usually do, so protecting the highlights and lifting the shadows yields the most natural merge. (§23.2)
20. False — Night mode's multi-second capture can't freeze a moving subject; the dancer would ghost/ smear even though the static background stacks cleanly. (§23.3)
21. The merge combines multiple exposures to capture a scene's full dynamic range (recovering both shadow and highlight detail no single frame could hold). The tone-map then compresses that wide range down to fit a display. The tone-map is what determines whether the result looks real (gentle mapping) or cooked (heavy, haloed, flat — the "HDR look"). (§23.2)
22. Any three of: stereo / two cameras (triangulating distance from two slightly-offset views, like two eyes); a dedicated depth sensor (projected IR dots or time-of-flight, a genuine 3-D measurement); machine-learned single-image depth (a trained model that guesses depth from one image). (§23.4)
23. Any four, in order, from: demosaic (reconstruct full color from the sensor's color mosaic) → stack/HDR/night merge (combine frames) → white balance + color boost (neutralize, then push toward pleasing "memory" colors — bluer skies, warmer skin) → local tone/contrast (open shadows, add punch) → sharpening (exaggerate edges for crispness) → noise reduction (smooth speckle, and fine texture with it) → segmentation-aware tweaks (detect sky/face/foliage/food and process each separately). (§23.5, Figure 23.6)
24. The empty alley held still, so all the night-mode frames aligned and averaged perfectly into a clean image. The friend moved during the multi-second capture, so her frames didn't reinforce — she ghosts. Fixes (any two): (a) have her hold completely still for the whole capture; (b) add light on her (step into a streetlight's pool, have someone shine a phone light) so the phone uses a shorter capture she can hold through; (c) shoot a single brighter exposure and accept some noise. The phone isn't defective — it's extraordinary on still subjects and helpless against motion. (§23.3)
25. Two pipeline behaviors: (1) computational color / memory-color boost is pushing the glaze toward a vivid teal that the real muted grey-green doesn't have; (2) aggressive sharpening plus noise reduction is over-crisping the obvious edges while smoothing away the subtle surface tool marks (the "watercolor" smoothing of fine texture). The fix: shoot RAW (or RAW+JPEG), set white balance yourself for accurate color, and develop with restrained sharpening and minimal noise reduction so the true color and the surface texture both survive — accuracy matters here, so don't trust the auto JPEG. (§23.5, §23.6, Figure 23.7)
Topics to review, by question
- Q1–2: Multi-frame capture and stacking; why averaging reduces noise → §23.1
- Q3–5, Q19, Q21: HDR, dynamic range, tone mapping, the "HDR look" → §23.2
- Q6–7, Q20, Q24: Night mode — alignment, stacking, and the motion weakness → §23.3
- Q8–11, Q17–18, Q22: Portrait mode, depth maps, synthetic bokeh, and its artifacts → §23.4
- Q12–13, Q23, Q25: Computational color, sharpening, noise reduction, semantic segmentation → §23.5
- Q14–16, Q24–25: The override decision; when computation lies; shooting RAW → §23.6