Unfamiliar Face Recognition Flashcards
UFR: Eyewitness Memory (Old Lineups)
The eyewitness/victim observe the criminal. Attend the police station, view a live line up of the suspect and several non-suspects, and are asked to pick out the criminal they observed during the attack. This is recognition from memory
UFR: Eyewitness Memory (VIPER)
Video Identification Parade Electronic Recording. People allow their face to be used in an electronic line-up. UFR: Eyewitness Memory (VIPER). The eyewitness/victim observe the criminal. Victim doesn’t have to be in same physical space/proximity to suspect, as it can be scary for the witness/victim. Provides police with greater range of non suspects for the line up
What is type 1 of unfamiliar face recognition?
Recognition from memory
What is type 2 of unfamiliar face recognition?
Unfamiliar Face Matching. Correctly deciding whether two unfamiliar face photos show the SAME PERSON or TWO DIFFERENT PEOPLE
Why focus on the face for criminal recognition? (why not voice, clothes…)
Pryke, Lindsay, Dysart & Dupuis (2004). Phase 1: Encoding: Witness to crime. Phase 2: Recognition (of the perpetrator) in different conditions (face, voice, body, clothes). Perpetrator Identification from
face photos was the most
accurate, evidence that using the face to identification over these other factors is most effective
What was the error rate in face identification in Pryke, Lindsay, Dysart & Dupuis (2004)
30%
Eyewitness Memory: Megreya & Burton (2008)
Participants observed a neutral “criminal” for 30 seconds, followed by a 10-lineup array 5 seconds later. They identified whether the “criminal” was present or absent. However unlike real-life scenarios, this process occurred almost immediately. Participants
picked the wrong person 30% of the time (average of both conditions). Criminal present: 38%, absent: 22%
What is a criticism of Megreya and Burton (2008)
Lack of ecological validity: shown line-up 5 seconds after “criminal” shown, in real life situations would be a bigger gap
Predicting Identification Accuracy
Morgan et al (2007). 53 U.S. Army personnel were exposed to an interrogation. High stress environment, social isolation training etc. 48 hours later, completed the Weschler Face Test, asked to identify one of their interrogators. Weschler Face Test: 35% Error Rates. Interrogator Identification: 38% Error Rates
What are issues with Morgan et al (2007).
Exact same image shown at learning and test, image recognition, not face recognition (issue with WFT). Only 33 participants, would want at least 100
Bruce et al (1999)
Matching custody photos to a CCTV image. The Task: Is the face from CCTV in the image array, if so, which one? 21% Error Rates! Non-Trivial Level of Error. Better than from memory, but still poor performance. When difference in pose, 32% error. Too many distractors (10)?
GFMT: The Glasgow Face Matching Test
(Burton, White, & McNeill (2010). 40 Trials. 2 Faces on Each Trial. 20 trials same person (match C), 20 trials different person (mismatch C). Even when we remove the distractor faces and go 1-1 matching, people incorrectly say that the CCTV suspect and the face on the right match (20% error)
GFMT: Heathrow Airport
20% Error Rates @ 100,000
people. 20,000 fraudsters with fake passports entering the UK. Unacceptable level of error in a national security context
GFMT: Other Race Effect
Caucasian observer/Caucasian face: 30% error. Caucasian observer/Egyptian faces: 48% error
What is a criticism of the GFMT (Burton, White, McNeill, 2010)
Lack of ecological validity: The faces in the task have a neutral expression and facing straight on
Ritchie et al (2023)
In unfamiliar face matching, people performed as poorly with unfamiliar faces as with money faces. Suggests unfamiliar faces are processed differently from familiar faces.
Realistic Courtroom Study: CCTV Image to Live Face Matching
Davis & Valentine (2009). Participants act as the Jury. They are shown a CCTV clip and have to decide whether or not the
defendant standing in the dock is the person they have seen in the CCTV footage. 19% Error Rates: no different to errors in photo only tests
Why is unfamiliar face recognition so difficult and so prone to error?
The Problem of Within Person Variability
Within Person Variability: Unfamiliar F’s
Jenkins, White, Van Montfort & Burton (2011). 20-40 photos, only 2 identities. On average, people think 7 different people. We don’t know how the face of an unfamiliar person varies
Hunnisett and Favelle (2021)
Presenting multiple varying images of an unfamiliar face improves matching accuracy. Suggests within-person variability can aid in unfamiliar face recognition.
