Chapter 26 — Exercises

Work these without looking back at the chapter first; then check yourself. Items marked † have full worked solutions in the answers appendix. There are no answers in this file. Mix of recall, applied reasoning, evidence interpretation, "spot the overstatement," ethics, and a cold-case extension.

A. Recall and definitions

  1. Define image forensics in one sentence, and name the three broad jobs this chapter says it does (measurement, authentication, and reading content).

  2. Define photogrammetry, and name the most defensible method for estimating a person's height from surveillance footage. In one sentence, say why that method "cancels" the camera's lens and angle.

  3. † State the one-sentence rule about what enhancement can and cannot do. Then explain why that rule makes "zoom-and-enhance" impossible.

  4. Define image authentication, and explain why this chapter calls it the field's "real frontier" while calling enhancement its "central myth."

  5. Define error level analysis (ELA) in your own words, and state plainly whether a bright region in an ELA map proves manipulation.

  6. Define a deepfake, and distinguish it from a cheapfake with one example of each.

  7. Define provenance and metadata, and name the common image-metadata standard (the one that can record camera model, time, and sometimes GPS).

  8. List four legitimate enhancement operations and, for each, name the existing information it makes easier to perceive.

B. Applied reasoning

  1. † A reverse-projection height estimate is reported as "approximately 179–185 cm." List three specific reasons the analyst reported a range rather than a single number, and name one real-world factor (about the person) that should make the range wider.

  2. Explain, step by step, what an investigator should do at a scene to preserve surveillance video so the lab can later work with it. Name the integrity check (from Chapter 25) that should be computed and why.

  3. A figure in footage is estimated at "approximately 165–171 cm" and the suspect is 188 cm. State the honest conclusion and name which of the book's honest verbs (exclude / consistent with / strongly supports) applies. Contrast this with a case where the estimate overlaps the suspect's height.

  4. † Frame averaging genuinely increases usable detail in some cases but not others. Explain when it works, when it does not, and why — using the idea that signal is consistent across frames while noise is not.

  5. A surveillance time stamp reads 11:38 p.m. Name three independent things you could check to decide whether to trust it, and explain why the stamp is a claim, not a fact.

  6. Distinguish automated facial recognition (database search) from facial comparison (examiner opinion). For each, state whether it is better understood as a lead generator or a proof, and why.

  7. A composite (pasted-in object) image is suspected. List three physics-or-content consistency checks an analyst could run that do not depend on metadata, and explain why each is hard for a forger to fake.

C. Evidence interpretation

  1. † Re-read Figure 26.1 ("Two enhancements of the same blur"). Version (A) makes two characters faintly readable; version (B) shows a crisp, fully legible plate. State which is admissible enhancement and which is fabrication, and explain the rule that separates them.

  2. An analyst projects a glowing ELA map and testifies, "The bright area proves this photo was edited." Identify the error, name two innocent features that routinely "light up" in ELA, and write the honest version of what ELA can support.

  3. A file's EXIF shows a "date taken" that is later than its "date modified." Explain why this internal inconsistency is a red flag, and state what it suggests without overstating.

  4. A defendant's alibi video has metadata that all agrees internally and matches the claimed device. Does this prove the video is authentic? Explain, using the asymmetry between "fails to exclude tampering" and "proves authenticity."

D. Spot the overstatement / junk-science alert

  1. † A prosecutor's slide reads: "Enhanced surveillance imagery positively identifies the defendant as the man buying gas cans." Identify two distinct overstatements in this sentence (one about enhancement, one about identification) and rewrite it honestly.

  2. An expert testifies that a machine-learning upscaler's output "shows the suspect's face that the camera captured." Name the specific scientific problem with calling the upscaled face "what the camera captured," and give the defensible description.

  3. A witness says, "It's on video, so it's proof he did it." Using the §26.1 caution about footage feeling like direct evidence and the §26.5 "liar's dividend," explain two different ways this sentence can mislead — one when the video is genuine, one in a world where fakes exist.

  4. A report concludes a video is "definitely not a deepfake" because a detector scored it 99% authentic. List the missing cautions (novel-method generalization, compression, error-rate validation) that should make a careful reader distrust the word "definitely."

E. Ethics and reasoning

  1. † An analyst is asked by the lead detective to "make the plate readable" from footage where only three of seven characters are legible at full resolution. Describe the ethical line between honest enhancement and producing a misleading image, and what the analyst should document and say instead.

  2. Two qualified analysts disagree about whether a region of an image was manipulated; one cites an ELA glow, the other attributes the glow to a high-contrast edge. Is this necessarily a scandal? Explain, and state which analyst is reasoning more defensibly and why.

  3. You are asked to testify that a deepfake-detection tool "guarantees" a confession video is genuine. Explain why you should decline given the field's current state (Chapter 5, Daubert; Chapter 6, foundational validity), and what you can honestly say instead.

  4. Automated facial-recognition systems have been shown to vary in accuracy across demographic groups. Using the bias theme (and previewing Chapter 31), explain why deploying such a tool as proof rather than lead is especially dangerous, and what safeguard the result requires.

F. Synthesis and validity spectrum

  1. † Place these four on the NAS 2009 / PCAST 2016 validity spectrum (strong → unsettled/contested), justifying each: documented reverse-projection photogrammetry; well-understood image enhancement (brightness/contrast/noise reduction); error level analysis as a standalone manipulation verdict; current deepfake detection.

  2. Explain how this chapter's metadata caution connects to Chapter 25's handling of digital evidence (hashing, write blockers, the fragility of metadata). Why is the same lesson appearing in two chapters?

  3. In one paragraph, explain how this chapter advances at least two of the book's four themes (exclusion over proof; the validity spectrum; cognitive bias; the CSI effect cutting both ways). Name which themes and how.

G. Cold-case extension

  1. † Cold Case. The gas-station CCTV shows "a person consistent with Keller buying gas cans." Write the entry you would add to the Mill Creek evidence log (Appendix I). State (a) the defensible inference at its true strength, (b) the honest verb, (c) at least three things this footage specifically does not establish, and (d) why "consistent with Keller" must not be logged as "identified Keller."

  2. Cold Case extension. Keller's alibi video has internally inconsistent metadata. Explain precisely what this does and does not do to the case: which way does it move the alibi, and why does undercutting the alibi not amount to proof that Keller was at the cabin? Use the metadata asymmetry from §26.6.

  3. Cold Case, integrative. The imagery "corroborates purchase and presence" but does not identify the killer. List two other evidence types already in the file (from earlier chapters) that, combined with the CCTV, strengthen the link between Keller and the crime — and state plainly why the footage alone cannot make that leap. (Think: the confirmed accelerant; the digital trail; soil.)

  4. Cold Case, method choice. Investigators want to strengthen the identification beyond "consistent with." Propose two honest, concrete steps (think: a better reference for photogrammetry; corroborating the figure's clothing or vehicle with other footage or evidence; provenance/authentication of the alibi file) and state what each would add — and what it still could not establish.

H. Short writing

  1. † In 150–200 words, explain to a juror why "enhancement" is mostly a television lie, what real enhancement can honestly do, and why a crisp, machine-generated face from a blur is more dangerous than an honest blur.

  2. In 150–200 words, explain why authentication and deepfake detection — not enhancement — are now the central problems of image forensics, and why a court should treat a confident "this is/ is not a deepfake" claim with caution.