EFIT-V Facial Recognition Software
Submitting InstitutionUniversity of Kent
Unit of AssessmentPhysics
Summary Impact TypeTechnological
Research Subject Area(s)
Information and Computing Sciences: Artificial Intelligence and Image Processing
Medical and Health Sciences: Neurosciences
Psychology and Cognitive Sciences: Psychology
Summary of the impact
Research conducted within the School of Physical Sciences (SPS) at the
University of Kent has led to the development and successful
commercialisation of facial identification software named EFIT-V.
First sold in 2007, this software is now used by more than 70 police
forces internationally and has revolutionized the way eyewitnesses
and victims of crime create computerised facial likenesses of offenders.
These images are circulated to police intelligence units, and the general
public, leading to the identification and arrests of offenders. Police
Identification rates have jumped from 5% to 55% as a result of this
software. With a current annual turnover exceeding £250K, which is
projected to reach £600K by 2015, Kent spinout company Visionmetric
has made significant impact with EFIT-V, and achieved a position
of commercial dominance in the UK, and around the world.
Before the introduction of EFIT-V in 2007, all existing
commercial computerised facial-composite systems employed a
feature-based approach, in which a witness is required to describe, and
then select the best matching individual facial features (eyes, nose,
mouth etc) from databases of stored images. Established research has
documented substantial weaknesses in this approach [Tanaka and
Farah J. Experimental Psychology 46A 225 (1993)]. In particular, facial
recognition by humans does not function through decomposition into
distinct features, but through a gestalt or holistic impression. Moreover,
the inability of humans to recall and describe faces accurately compared
to their superior ability to recognise faces is well established. Thus
approaches emphasising the need for recall and description do not match
human cognitive processes, and are inherently disadvantaged. Studies
conducted prior to the introduction of EFIT-V suggested that the
operational identification rate of police forces using
feature-based systems worldwide was only 5%. [McQuiston-Surrett
et.al Psychology, Crime and Law 12 505 (2006)].
Research within SPS underpinning EFIT-V commenced in 1997 and
continues to this day. Initial published work demonstrated the utility of
a set of matched 2D polynomials to estimate optically
distorted wavefronts, and to represent and automatically classify
impact craters efficiently (see references to research section). A key
insight was that quasi-stochastic processes (impact craters being one
example) are numerous and the extension of these methods to the stochastic
pattern class of human faces yielded an extremely efficient parametric
representation. Calculation of a Karhunen-Loeve basis set enabled a
recognisable, near photographic quality facial image to be stored in just
50 bytes. Based on this, Dr. Solomon (SPS since 1994, Lecturer now
Reader) led the development of both a prototype system for encoding the
owner's face on a credit card's magnetic strip (reported in New Scientist
Aug 29th 1998 and on Tomorrow's World, 1998) and FaceMail (winner of the
UK and European stages of the EU-sponsored DICON (digital information
contents) 2000 competition) (see section 5 for link to further details).
FaceMail worked by attaching an encoded picture of the sender to the
email-header providing a visual impression of the sender to aid
prioritisation and handling of spam messages.
In 2001, Solomon led a research team to begin exploring methods of
computerised facial synthesis that combined calculated sets of
Karhunen-Loeve functions for the shape and texture of the face with an
interactive search algorithm that enabled witnesses to converge on the
required identity by manipulating the underlying Karhunen-Loeve
coefficients. Dr Gibson (SPS since 1994, UG, PhD, Postdoc and Lecturer)
developed suitable statistical models across all varieties of human facial
appearance. Research students made significant contributions, including
A.Pallares-Bejarano (Kent PhD 2002-2006) who developed the most efficient
stochastic/evolutionary search algorithm, M.Maylin (Kent PhD 2002-2006,
PDRA 2006-2009) resolved issues of software implementation, computational
efficiency and speed, and C.Scandrett (Kent PhD 2003-2007) who made
significant contributions to automated age progression, which was
ultimately employed in the commercial software.
Four successive EPSRC grants (Solomon, PI, 1-4 in Section 3) were secured
to develop the underlying science and extend these ideas commercially. All
four were rated Outstanding for potential benefits to society. Commercial
potential was recognised early and two UK patents were lodged in 2005. The
basic elements of the developed experimental system were published in 2003
(reference 2). Two key advancements that were subsequently implemented in
the commercial EFIT-V system were automated caricaturing [Gibson
et al, Behaviour Reseach 37 170 (2005)] and automated age
progression [Scandrett et.al Pattern Recogniton Letters, 27 1776
A comprehensive account of the underpinning science has recently been
published. [Solomon et al Applied Soft Computing 13 3298 (2012)].
Key insights underpinning the impact of EFIT-V were:
- Statistical techniques researched by the Kent group for treating
stochastic processes (e.g. impact craters and atmospherically distorted
wavefronts) could be extended to human faces with unparalleled
- The resulting compact representation provided a mathematical model of
facial appearance that was better matched to human face recognition and
cognitive processes than the crude feature based model employed in
earlier commercial systems.
- The model allowed eyewitnesses new, flexible ways to alter facial
appearance (e.g. fully automatic age progression) and alteration of
subjective attributes (e.g. weight loss/gain, hostility, health).
- A viable stochastic search algorithm could be defined enabling the
underlying model to be manipulated easily by human observers without
technical or specialist knowledge to achieve satisfactory likenesses.
References to the research
Four EPSRC research grants supported the research carried out. All 4 were
rated Outstanding for potential benefits to society:
1. EPSRC GR/S06738/01 "Synthesis of facial composites for improved
suspect identification" Principal Investigator, Dr C.J. Solomon. April 1
2003 — March 31 2005". Amount: £107,414
2. EPSRC GR/ S98504/01 "SWISS — Significant Witness Identification of
Suspects System" Principal Investigator, Dr C.J. Solomon. Oct 1 2004 — Sep
30 2006". Amount: £106,161
3. EPSRC EP/D040973/01: "Statistically rigorous age progression for the
identification of missing persons" Principal Investigator, Dr C.J.
Solomon. Jun 1 2006 — Aug 31 2009. Amount: £177,638
4. EPSRC GR/S57839/01: "Modal Analysis For Exact Forensic Comparison of
Handwritten Documents" Principal Investigator, Dr C.J. Solomon. Feb 1 2004
— Jan 31 2005. Amount: £58,714
1. "Generation of facial composites", PCT/GB2005/002780. 25/7/2005.
Inventors: C.J. Solomon, S.J. Gibson and A. Pallares-Bejarano
2. "Plausible Ageing of the human face", PCT/GB2005/002669. 06/7/2005
Inventor: C.J. Solomon.
Publication in the open literature was consciously restricted for reasons
of commercial confidentiality. All listed authors below are either Kent
academics or Ph.D students. Publications ,  and  marked with a *
best indicate the quality of the underpinning research.
1* L. Kay, A. Podoleanu, M. Seeger and C.J. Solomon, "A new approach to
the measurement and analysis of impact craters" Int. Journal Impact. Eng.,
Vol.19, No. 8, pp739-753 (1997). http://dx.doi.org/10.1016/S0734-743X(96)00054-1.
The use of matched polynomial sets to describe and classify impact
2* S. J. Gibson, C. J. Solomon and A. Pallares-Bejarano, "Synthesis of
photographic quality facial composites using evolutionary algorithms",
Proceedings of the British Machine Vision Conference 2003, Vol 1,
Cited 41 times according to Google scholar, this is the first publication
describing the overall system concepts and prototype EFIT-V system.
3* S. J. Gibson, C. J. Solomon and A. Pallares-Bejarano, "Non-linear,
near photo-realistic caricatures using a parametric facial appearance
model" Behaviour Research: Methods, Instrumentation and Computing, 37(2),
170-181, (2005). http://dx.doi.org/10.3758/BF03206412.
This paper describes our use of the facial appearance model based on PCA
to automatically generate photo-quality caricatures.
4. D.Wallis, M.J. Burchell, A.C. Cook, C.J. Solomon and N. McBride
"Azimuthal impact directions from oblique impact crater morphology", Mon.
Not.. R. Astron. Soc, 359, 1137-1149, 2005, http://dx.doi.org/10.1111/j.1365-2966.2005.08978.x.
The most mature expression of our use of matched polynomial sets to
describe and classify craters according to kinematic impact.
Details of the impact
Older facial composite systems perform poorly as they rely on the memory
of victims or eyewitnesses under the guidance of trained police officers.
However, in the absence of forensic evidence, facial composites often
provide the only means of establishing leads in police investigations. The
introduction of EFIT-V has revolutionised the production of
realistic facial images of criminals in the UK and worldwide. Two
representative testimonials are given below:
"We decided after these tests [initial trials of EFIT-V] to have a
"leap of faith" and go with EFIT-V... since then our naming rate has
increased dramatically to around 55%" West Yorkshire Police 2009.
"I thought it would be appropriate to let you know as soon as possible
about the new product [EFIT-V]. Put simply, I love it...This has led to
6 cases so far getting "near as damn-it" likenesses." Metropolitan
Police (New Scotland Yard) 2010.
Beneficiaries and Impact
The impact of EFIT-V is summarised in the following points:
(1) The use of EFIT-V is quicker and better matched to the natural
cognitive processes of witnesses, so officers can produce more accurate
likenesses rapidly. As a result, it has become the new industrial standard
used in 85% of the 51 police forces in the UK.
(2) The technology has led to a major increase in reported
identification rates from police forces using the system. The
feature based approach described above has historical identification rates
at 5% [Tanaka and Farah J. Experimental Psychology 225 (1993)], whereas
rates as high as 55% are reported with the use of EFIT-V as
commented on by West Yorkshire police. Overall, since its release in early
2008, licensees have seen increases of more than 100% in useful
intelligence (as reported in the Investigative Practice Journal of Police
Professionals, the largest circulation police weekly). An extended study
of its performance in the field involving more than 1000 interviews
resulted in an exceptional 40% naming rate [Driver and Rowbotham, E-FIT
user conference 2009]. Construction is also up to 100% quicker than for
traditional, feature-based systems.
(3) It has spawned a successful spinout company, Visionmetric,
which has exported the technology to countries around the world including
Sweden, Singapore, Jamaica, Canada, Ghana, Colombia, Botswana, Malta,
Chile, Macedonia, Canada, Slovenia, South Africa, and the USA. The
company's current turnover exceeds £250K p.a. and is projected to reach
£600K p.a. by 2015. Visionmetric's directors, Solomon and Gibson, are Kent
SPS academics and the prime contributors to the underpinning research.
(4) It has played a key role in apprehension of hundreds of
criminals, many involving very serious crimes.
Sources to corroborate the impact
The following persons may be contacted to corroborate the reported
improvements in efficiency and identification rates in their particular
 Detective Sergeant, Crime Performance and Strategy Unit, Wiltshire
Police [Contact 1] can corroborate the reported improvements in
efficiency and identification rates relevant to the Wiltshire Police
 Facial Imaging Officer, formerly of West Yorkshire Police, now
retired [Contact 2] can corroborate the reported improvements in
efficiency and identification rates
 Detective Constable, New Scotland Yard, Metropolitan Police [Contact
3] can corroborate the reported improvements in efficiency and
identification rates for the Metropolitan Police
 Facial Identification Officer, Q-Block, Essex Police Headquarters
[Contact 4] can corroborate the reported improvements in efficiency and
identification rates within the Essex Police area of responsibility
The company website  Visionmetric (http://www.visionmetric.com/)
and a website specifically designed to store REF 2014 related information
- that the EFIT-V software is based on the underpinning research
undertaken at Kent
- that both the underpinning research and its commercial realisation
have significantly impacted on police practice and received considerable
- of the international user base of EFIT-V supplied by
distributers across 18 countries
These websites also provides direct access to publically available
material in the form of popular science and general interest articles, web
pages, customer testimonials and financial statements of the company.