Chapter 26 — Self-Check Quiz

24 questions: multiple choice and short answer. Try them closed-book. The answer key is in the collapsed block at the bottom.

Multiple choice

1. Image forensics primarily concerns: - A. Making any blurry footage perfectly clear - B. The analysis, measurement, and authentication of images and video for legal purposes - C. Editing photos to look better in court - D. Generating synthetic suspects

2. The most defensible method for estimating a person's height from surveillance video is: - A. Guessing from nearby objects - B. Reverse projection (returning to the scene and imaging a known reference at the same stand point with the same camera) - C. Reading the height off the camera's metadata - D. Asking the suspect

3. Reverse projection works because: - A. It assumes the camera's specifications from the manufacturer - B. The known reference and the unknown person share the same camera angle, lens, and perspective, so those distortions cancel - C. It uses artificial intelligence to add detail - D. It ignores perspective entirely

4. The one true rule about enhancement is: - A. It can recover any detail with enough processing - B. It makes information already present easier to perceive but cannot add detail the sensor never recorded - C. It always improves identification - D. It creates a license plate from any blur

5. A machine-learning "super-resolution" image that renders a crisp plate from a six-pixel blur is: - A. Reliable evidence of the plate number - B. Hallucinated detail reflecting the model's training data, not what the camera captured — not evidence - C. The gold standard of enhancement - D. Required by Daubert

6. Frame averaging genuinely increases usable detail only when: - A. The subject is moving quickly - B. The scene/subject is static across the combined frames (signal consistent, noise random) - C. The file has been re-saved many times - D. The camera was replaced between frames

7. A surveillance time stamp should be treated as: - A. An unquestionable fact - B. A claim made by the recording device, to be verified against settings/metadata/a known event - C. Always correct if the camera is digital - D. Irrelevant to an investigation

8. Error level analysis (ELA): - A. Definitively proves whether an image was edited - B. Is a screening hint (edited regions may compress differently) that can be confounded by edges, text, and texture — not a standalone verdict - C. Works perfectly on screenshots and all formats - D. Identifies the camera that took the photo

9. A bright region in an ELA map is commonly produced innocently by: - A. Only by manipulation - B. Sharp edges, high-contrast boundaries, overlaid text, and fine texture - C. The camera's serial number - D. The presence of a deepfake

10. A deepfake is best defined as: - A. Any photograph - B. Synthetic or manipulated media generated/altered by machine-learning techniques to depict a real person doing/saying something they did not - C. A blurry video - D. A photo with no metadata

11. A "cheapfake" differs from a deepfake in that it: - A. Is always more convincing - B. Uses simple, non-AI edits (slowing footage, mislabeling, cropping) but can still do real damage - C. Cannot mislead anyone - D. Requires a neural network

12. The "liar's dividend" refers to: - A. Money paid to expert witnesses - B. The way the mere existence of convincing fakes lets people dismiss genuine evidence as "probably a deepfake" - C. A type of compression artifact - D. A photogrammetry error

13. Provenance is: - A. The visible content of an image - B. The documented origin and history of a file (source device, capture time/place, edit history) - C. The brightness of an image - D. A deepfake-generation method

14. Regarding metadata such as EXIF, the chapter's recurring caution is that: - A. It is always reliable and proves authenticity - B. It is fragile and forgeable — easily stripped or edited — so its presence must be authenticated and its absence proves nothing - C. It cannot record time or location - D. It is impossible to remove

15. Photogrammetric height evidence is most honest and powerful when it is used to: - A. Identify a specific suspect - B. Exclude a suspect whose height clearly falls outside the estimated range - C. Prove the time of the crime - D. Authenticate metadata

16. Compared to documented reverse-projection photogrammetry, current deepfake detection sits on the validity spectrum as: - A. Equally validated and quantified - B. A real but unsettled, fast-changing frontier without well-characterized error rates on novel fakes - C. Fully discredited like bite marks - D. Having no investigative value at all

17. In the cold case, the gas-station CCTV most defensibly establishes: - A. That Keller, and no other person, bought the gas cans - B. The exact time of the fire - C. That a person consistent with Keller bought gas cans (corroboration, not identification) - D. That Keller killed Diallo

18. The most promising long-run defense against fabricated media is: - A. Better after-the-fact deepfake detection alone - B. Cryptographic content-provenance (signing media at capture and tracking edits), so genuine media can prove its lineage - C. Banning all cameras - D. Ignoring the problem

Short answer

19. In two sentences, explain why a phone video of a security monitor is not the evidence you want, and what is.

20. Name three sources of error in a reverse-projection height estimate, and for each, state in one phrase how it can distort the result.

21. Write the one-sentence rule that separates legitimate enhancement from fabrication, and give one example of each side of the line.

22. Explain why ELA is useful as a screening tool but dangerous as a standalone verdict. Name one feature that confounds it.

23. A defendant's alibi video has internally inconsistent metadata. Explain what this does to the alibi and why it does not, by itself, prove the defendant was elsewhere guilty — use the metadata asymmetry.

24. State, in one sentence each, what a photogrammetric height match and a deepfake-detector "authentic" result can each honestly support on the stand — and what neither can claim.

Answer key (click to expand) **Multiple choice:** 1-B · 2-B · 3-B · 4-B · 5-B · 6-B · 7-B · 8-B · 9-B · 10-B · 11-B · 12-B · 13-B · 14-B · 15-B · 16-B · 17-C · 18-B **Short answer (model points):** **19.** A phone video of a monitor is a copy of a copy: it is re-compressed (degrading detail), stripped of the original's embedded metadata, and unverifiable as to integrity. The evidence you want is the *native* file exported from the DVR at full resolution with its metadata intact, hashed (Chapter 25) so its integrity can be proved. **20.** Any three, e.g.: (a) **stand point** — if the reference is placed where the person did not actually stand, perspective shifts the apparent size; (b) **posture/footwear/headwear** — slouching, leaning, shoes, and hats change apparent height by unknown amounts; (c) **camera change** — if the camera was moved/replaced/refocused, the projection no longer matches; (d) **image quality** — blur/low resolution makes head-top and floor-line ambiguous, and each pixel of ambiguity is centimeters at distance. **21.** **Rule:** enhancement reveals information already present in the image; it never adds detail the sensor did not record. **Legitimate:** brightening shadows so an already-captured face becomes visible. **Fabrication:** a neural upscaler generating a crisp, legible plate from a few pixels — the characters are guessed, not captured. **22.** ELA can draw the eye to a region that compresses differently from its surroundings — a possible edit — making it a quick screening hint. It is dangerous alone because innocent features (sharp **edges**, high-contrast boundaries, overlaid **text**, fine texture) routinely "light up," and ELA fails on multiply re-saved images, non-JPEG formats, and screenshots; a bright region is a prompt to look closer with *other* methods, not proof. **23.** Internal inconsistency (e.g., a "date taken" later than "date modified," contradictory timestamps, a software-editor field) is a strong red flag that the file was altered, which *undercuts* the alibi's claim to fix the defendant's location/time. But by the metadata asymmetry, inconsistency can *undercut* a claim while consistency only *fails to exclude* tampering — so undercutting the alibi removes a point in the defendant's favor; it does not place him at the crime or prove guilt. **24.** A **photogrammetric height match** honestly supports that the figure is *consistent with* the suspect (and can *exclude* a suspect of clearly different height); it cannot *identify* the suspect to the exclusion of all others of similar stature. A **deepfake-detector "authentic" result** can support that no detected synthetic signatures were found; it cannot *prove* the media is genuine, because detectors degrade on novel fakes and on compressed/re-shared media and their error rates on such cases are not well characterized.