Generating face images from brain waves

How are suspects identified and located when the only lead in the investigation is an eyewitness’ memory of events? In these situations Police will often create a pictorial likeness of the suspect called a Facial Composite (a.k.a. E-FIT, PhotoFIT or Identikit). Previously, composite images have been created from a verbal description of the suspect, provided by a witness. This is a lengthy process that can be prone to misinterpretation which adversely affects accuracy and therefore also diminishes the chance of locating the suspect. Wouldn’t it be great if we could overcome these issues by simply reading the witness’ mind? That’s what SPS spin-out company Visionmetric ltd aim to achieve with their EEG-FIT research and development project.

The EEG-FIT project aims to exploit the enhanced (so-called Fringe P3) EEG brain wave response to facial images that bear resemblance to a criminal suspect and, in this way, generate facial composite images more accurately and rapidly.

Key people: Dr Chris Solomon, Dr Stuart Gibson & Prof Howard Bowman.

Trust embraces artificial intelligence to improve patient care

Kent Researchers, Bai & Gibson, develop AI based Clinical Support Software to aid Ophthalmologists.

Senior lecturer Dr Stuart Gibson said: “AI has completely revolutionised the way we approach computer vision research. Our team has considerable experience in this area, having previously developed AI for facial identification, detection of objects concealed in postal items and the identification of unknown substances.”

“The primary motivation for our work is to have a positive impact on society. Our project with Nishal Patel and the Trust has the potential to significantly improve patient care.”


Biomedical Sciences Student’s Day

Fangliang Bai (SPS Postgrad) delivers review presentation on  Clinical Support Systems underpinned by AI. 7th November 2019, Biomedical Sciences Students Day at East Kent Urology, Kent and Canterbury Hospital.

Deep convolutional neural networks for Raman spectrum recognition

Love it or loath it, artificial intelligence has revolutionised computer vision and continues to generate great interest in many areas of science. In our 2017 Analyst paper we show that a 1D convolutional neural network is an effective tool for the classification of unknown substances.  Our method, developed for vibrational spectroscopy,  removes the necessity for complicated signal preprocessing  and therefore allows rapid and automated identification of substances. We are currently working with industry to develop this technology further for application in large scale chemical manufacturing.

Testing Newly Developed Sub Sea Tracking Solution

I’m currently an Honorary Researcher at Cranfield University and co-supervising Wojciech Lebkowski (Cranfield, CSTE MSc programme ) in collaboration with Seismic Stuff. Read about the project here.

Research projects

(1)  Forensic Image Processing
In the last decade, affordable digital camera technology has become widely available, resulting in the proliferation of digital images. The creation, modification and distribution of certain photographic materials are controlled by UK law (e.g. Protection of Children Act & Counter Terrorism Act). Two procedures that are particularly useful in the detection of image based crime are (i) matching a digital image to its source camera and (ii) identifying alterations to digital photographs. The aim of this project is to develop algorithms that support both of these tasks under the challenging conditions of file sharing over mobile phone networks and on social networking websites. The project researchers will be working with The Public Protection Unit, Kent Police.

Researchers: Susan Welford & Dr Stuart Gibson

Keywords: digital image forensics, Photo Response Non Uniformity (PRNU), image denoising, image compression 


(2) Optimization of Image Data using Interactive Evolutionary Computation

Interpretation of image data by human analysts is prevalent in many fields including medical imaging, astronomical imaging, and the biological sciences. An important, and frequently adopted, preliminary step in the interpretation process is to enhance the image of interest using standard image processing techniques thereby assisting in the objective (e.g. a medical diagnosis). The optimal image processing technique and associated parameter settings can depend on a number of factors such as the extent and type of any degradation, the particular image, or the purpose of the image. The specific nature of these factors may not be known in advance. In such circumstances it is reasonable to assume that a human analyst should be able to adapt the image enhancement process to suit a specific goal. The aim of this project is to develop interactive evolutionary algorithms that facilitate image enhancement. Specific areas of interest include image denoising, interactive morphing of facial appearance for effective composite construction and enhancement of astronomical image data.

Researchers: Mr Joseph Mist & Dr Stuart Gibson

Keywords: interactive evolutionary computation, evolutionary strategy, facial composite construction, image denoising


(3) Classification of Raman Spectra for the Identification of Counterfeit Goods

Raman Spectrometry is a non-destructive technique that can be used to analyse the composition of materials. Improvements in the accuracy and speed of laboratory based equipment, and introduction of portable devices, make Raman Spectrometry an appealing analytical technique for forensic applications, to include the detection of counterfeit goods. This project focuses on the analysis of spectra including noise reduction, automatic baseline fitting, data reduction and classification.

Researchers: Dr Stuart Gibson and Prof Michael Went

Keywords: raman spectrometry, Principal Components Analysis (PCA), chemometrics, machine intelligence, Independent Component Analysis (ICA), classification and kernel methods


(4) Personal Identification by Pictures (PIPs)

Personal Identification Numbers (PINs) are an established method for authentication that is often used in conjunction with tokens such as bank cards to access funds or services. PINs are also used to lock mobile phones preventing unauthorised use in the case of theft. Unfortunately, the human mind is poorly equipped to remember and recall numerical strings. In this project we investigate the use of rich (and hence memorable) visual stimuli as an alternative to PINs. We call this method Personal Identification by Pictures (PIPs).

Researchers: Mr Tat-Him Wai, Dr Christopher Solomon and
Dr Stuart Gibson

Keywords: personal identification, access security, cognitive psychology

(5) 12ft.lbs – An Arbitrary Value or a well-considered Limit of Firearm Lethality?
The 1968 Firearms Act (as amended) states that: “A firearm is a lethal barrelled weapon of any description, from which any shot, bullet, or other missile can be discharged.”
Bearing the above definition in mind, another part of the Firearms Act states that air rifles and air pistols with muzzle energies below 12ft.lbs and 6ft.lbs respectively are not legally firearms.  This means that weapons of these types with muzzle energies greater than the stated values instantly become firearms under the definition of being “lethal barrelled weapons” by the UK government.  This formally sets a limit of lethality in terms of muzzle energy.
This project would look at the current legal limits placed on the muzzle energy of air rifles of 12ft.lbs and 6ft.lbs for air pistols to assess whether they are suitable values.  This will be achieved through penetration experiments into physical simulant models of various parts of the human body to help consider if the above-stated values under UK firearms law are representative of being on the “boundary of lethality”.  If (as expected) issues are raised, the project will look into a more suitable system of representing lethality for classification of weapons under firearms law.

Researcher: Dr Christopher Shepherd

Keywords: limit of firearm lethality, muzzle energy, physical simulant models

Kent researchers awarded major EPSRC grant

Improving cyber security using realistic synthetic face generation

The grant application outlines a novel programme of research that questions the uniqueness of facial identity and investigates the use of computer generated face imagery in the area of cyber security. The popularity of the human face as a biometric remains strong despite the introduction of many competing modalities. People are accustomed to being identified by their facial appearance whereas other biometrics such as fingerprints and iris recognition feel more invasive. A programme of research that investigates the concept of identity that is highly relevant to cyber security is proposed. In addition we will develop a novel cyber security application based on facial identity and evaluate its practical security level.

Synthetic face generated by Kent team's model
Synthetic face generated by Kent team’s model

Work of this nature has relevance beyond the scope of the project. For example, border control officers routinely verify a person’s identity using passport photos but what is the fundamental limit on the ability to achieve this task reliably?

Principal Investigator Stuart Gibson (SPS), Co-Investigators: Julio Hernandez-Castro (School of Computing) & Chris Solomon (SPS).

The award coincides with Kent’s recent success in securing an Excellence in Cyber Security Research award, supported by GCHQ and EPSRC.

ICPS 2015 Amsterdam

People encounter more faces than ever in a modern world. This symposium investigates the boundaries of face recognition, including research on how many faces can be perceived at a time, imaging data on developmental face blindness, the existence of human face experts, and technology to increase face recognition performance.

Left to right: Stuart Gibson, Volker Thoma, Ashok Jansari and Davide Rivolta


Face-Specific Capacity limits
Volker Thoma
University of East London, United Kingdom

fMRI Correlates of Developmental Prosopagnosia
Davide Rivolta
University of East London, United Kingdom

Superior Performance in Face Recognition
Ashok Jansari
Goldsmiths University of London, United Kingdom

Interactive Evolutionary Generation of Facial Composites
Stuart J. Gibson
University of Kent, United Kingdom

Co-Author: Christopher J. Solomon Dr., University of Kent

SEPnet GRADnet Winter School 2015

The 2015 SEPnet Winter School for Physics postgraduate students took place at the Culham Science Centre (home to JET) in the week beginning 2nd February 2015. There was an unexpected flurry of snow on the Monday evening.


Back by popular demand, the Patterns and Security workshop was delivered by Tony Mansfield (Industrial Speaker, NPL) and Stuart Gibson (Academic Lead, University of Kent).   Presentations on the cutting edge biometric technology and an introduction to pattern recognition, were followed by a hands on activity in which students performed a hierarchical clustering of objects.

Forging a photo is easy, but how do you spot a fake?

Check out my article on The Conversation
Thanks to Michael Parker (TheConversation) for inviting me to submit and editing the article.