Matching views of unfamiliar people: how useful are passports and CCTV?

1
Matching views of unfamiliar people how useful
are passports and CCTV?
2
Faces are a primary means of identification Pass
ports and identity cards CCTV images for
identification and prosecution
3
CCTV cameras are widespread No good data on how
many exist! McCahill and Norris (2003) 4.2
million cameras in U.K. "A CCTV camera for every
14 citizens. (David Davis, resignation
statement, 12th June 2008) C4 FactCheck highly
questionable estimate. Gerrard and Thompson
(2008) 1.85 million (1.7 million privately
owned). BigBrotherWatch(2012) at least 51,600
local authority-controlled CCTV cameras, at a
cost of 515 million from 2007-2011.
4
Ways in which CCTV might reduce crime (Tilley
1993) Increases chances that offenders will be
caught, stopped or punished and therefore
deterred. Deters potential offenders who will
not want to be observed by CCTV operators or have
evidence against them captured on
camera. Encourages more "natural surveillance",
as more people use the area covered by CCTV. This
may deter offenders. Facilitates effective
deployment of security staff and police to
locations where suspicious behaviour is occurring
- either catching or deterring offenders. CCTV
cameras and signs show people are taking crime
seriously, and thus offenders may be
deterred. Publicity about CCTV may encourage
potential victims to be more security
conscious. Those who are more security-minded
may increase their use of areas with CCTV.
5
Few good data on CCTVs actual effectiveness As
a deterrent to criminals- Welsh and Farrington
(2002) Meta-analysis of 18 studies half showed
benefits, half did not. Gill and Spriggs
(2005) Evaluated 13 CCTV systems. Only 2
associated with a signifcant reduction in crime
levels. Significant reduction in public anxiety
in 3 areas. As prosecution evidence - Attorney
Generals Reference (2003) if "sufficiently
clear", CCTV footage can be used as sole
identification evidence.
6
Matching different views of an unfamiliar face is
highly error-prone Kemp, Towell and Pike
(1999) Supermarket credit-card study. Cashiers
falsely accepted over 50 of fraudulent cards,
and falsely rejected over 10 of legitimate
ones. Burton, Wilson, Cowan and Bruce
(1999) Subjects judged whether faces seen in
high-quality photos had been seen before in video
clips. Subjects personally familiar with the
targets performed well subjects unfamiliar with
them performed poorly. Henderson, Bruce and
Burton (2001) Subjects judged whether each of
two "robbers" was shown in video stills - poor
performance, with high false identification
rates.
7
Megreya and Burton (2008) Unfamiliar-face
matching tasks (using photos or real people,
matched to high-quality photographs).
Expt 3 simultaneous, no distractors (passport/ID
card matching)
Expt 2 simultaneous
Expt 1 short delay
15 error rate, plus bias to say "same" when live
TP 70 hits TA 35 false positives
TP 60 hits TA 20 false positives
Little difference between "live" and photo
targets unfamiliar face encoding is poor.
8
Can jurors reliably match images to
suspects? Davis and Valentine (2008) 3
experiments in which a live person was shown
together with high-quality video. Match between
person and video on 50 of trials, mismatch on
remainder. Highly error-prone and no better than
matching to stills. Worse with disguise, or with
year's delay between filming and test.
Wide variation between videos - overall 22 of
participants made false negative decisions (i.e.
saying "different" when actually "same"). 17
of participants made false positive decisions
(saying "same" when actually different).
37 failed to identify A as man in video 44
misidentified B as man in video
B
A
9
Techniques for "objectively" establishing
identity matches facial mapping
Aim - to determine whether two images show the
same person or different people. Normally used
where the CCTV image is too poor quality for
jurors to decide easily for themselves. Various
"Facial Mapping" techniques Measurements from
images (grid overlay, angular measurements,
etc.) Chimeric faces (two image halves
juxtaposed) Video wipe techniques (one image
progressively overlays another).
10
Techniques for establishing identity matches
facial mapping by overlaying a grid
11
Bromby Scale (2003) used by Facial Mappers to
express their opinion on whether a match has been
established
Merely a spuriously-quantitative expression of
subjective belief, but - Atkins vs Queen
(2009) does not question the technique's
scientific basis rules that the scale can be
used to express the Mapper's opinions.
12
Techniques for establishing identity
matches photo-anthropometry - angular
measurements Kleinberg, Vanezis and Burton
(2007)
Calculated different measurements from target
faces - six proportion indices (e.g. AA/BB)
and six angle values (e.g. AB/BB). Compared
each target video image against an array of 10
photos.
13
Kleinberg, Vanezis and Burton (2007)
All measures equally poor, even in these ideal
conditions (in real life, changes in viewpoint,
lighting and expression, poor quality video
images). Anthropometry is an unreliable
technique for matching images.
14
Davis, Valentine and Davis (2010)
Photo-anthropometry tested under ideal
conditions DigitalFace software used to obtain
37 linear and 25 angular measurements from two
high-quality images from each of 70 faces
(full-face and profile). Compared these to
measurements on separate probe images. Identificat
ion verification was unreliable unless multiple
distance and angular measurements from both views
were included.
15
Biases evoked by techniques used in court by
facial mappers (1) Strathie, McNeill and White
(2012) Matching is less accurate with aligned
chimeras than with full-face images aligned
chimeras evoke holistic processing and a bias
towards same responses.
16
Biases evoked by techniques used in court by
facial mappers (2) Strathie and McNeill
(2012) Matching is less accurate with video
"wipes" than with full-face images bias towards
same responses. (A possibility noted by Vanezis
and Brierley, 1996).
17
Conclusions Technological limitations of
CCTV Wide-angle lenses distort images Cameras
often positioned high, giving foreshortened
views Poor lighting conditions Poor image
resolution, especially with data-compression
images inherently lack information Criminals use
disguises Psychological limitations of
unfamiliar face-matching Even under ideal
conditions, unfamiliar face-matching is
poor. Dangers of facial mapping
techniques Facial mapping techniques are
unreliable and should not be admissible as
evidence in court (McNeill and Strathie
2012). Facial mapping evidence reduces jurors
(already poor) ability to match CCTV images to
defendants.