The key objective of the H-Unique project is to address the extremely important question of whether the human hand is unique.
That this might be so, is predicated on the likely fact that the human hand presents anatomical features that extend across more than six different aetiologies (causes or origins of variation). The likelihood of any two individuals showing matched features that arise from different aetiologies is highly unlikely, but we don’t yet know how unlikely.
“The hand retains and displays many anatomical differences due to our genetics, development, environment or even accidents so each person’s hands are different. Now for the first time, researchers will analyse all the factors that make a hand truly unique so we can understand and use them reliably as evidence to identify individuals.”
Professor Dame Sue Black on the H-unique project
Given that the closest comparators would be identical twins, it is well known that they possess different fingerprints, knuckle skin creases, superficial vein patterns, pigment patterns and scars. Therefore, even when the genetic influencers are identical, other extraneous features interact to cause detectable variation that could be better exploited
Sue Black
In 2006 our team was the first to take anatomical feature identification for the hand into the UK courtroom. The case was based on a young girl who alleged that her father was sexually abusing her. Leaving her skype camera running at night, she recorded the abuse although only the hand and forearm of the perpetrator was visible. As the evidence was novel, the Judge called a voir dire and ultimately decided that based on knowledge from anatomy, the methodology was sound and therefore admissible. This was a first for the UK courts [14] but it led to over 150 cases across the UK and further afield including INTERPOL, EUROPOL Norway, Spain, Thailand, Indonesia, Australia, Canada and the USA via the FBI [15-18]. In total, we have secured over 35 life sentences and over 200 years of incarceration for sexual crimes against children. We have calculated, that of the cases that we accept, over 80% result in a change of plea which is an enormous fiscal saving regarding court costs but, more importantly, it removes the obligation on the victim to give evidence in court, significantly reducing their anxiety and stress. This often leads to unacceptably high false acceptance and false rejection rates. This is a classic example of a problem in the real world, leading to a new and emerging research field.
In principle, this research will aim to evaluate the level of human variation that is both visible and can be extracted automatically, from digital images of the adult human hand. It will achieve this by constructing ground truth databases and from these, design algorithms that will extract individuating anatomical features from images and permit comparison across a large ground truth dataset to establish levels of variation and the certainty of being able to secure a match. We will assess the effects of dynamic hand pose and altered image quality on the ability to locate and extract information accurately and thereby construct a hierarchy of preferred features which will inform practice. Through a combinatorial approach, we will retro-engineer a multimodal suite of intelligent biometric filters that will allow large collections of images to be searched automatically to find a match for a single hand. Our ultimate research question to be answered is ‘Is the adult human hand unique’? Because if it is, then through combinatorial biometrics we will have shifted the frontier for biometrics in the security sector, forensic investigation and the identification sciences.
A key milestone within H-Unique is to how unique each anatomical feature in the hand is by examining the hands of at least 5,000 people. To do this we will construct a database with the help of volunteer contributors, which we believe will constitute the largest database of hand images in the world.
It typically takes 15-30 minutes to manually analyse each photograph. With H-Unique, we are aiming to analyse images of the hands of 5000+ participants in 5 different poses. This would take approximately 750,000 – 1,500,000 minutes or 1.4 – 2.8 years working flat out without a break. So we need to automate the process.
Key questions to be addressed
There are five work packages in this project and each poses a relevant key hypothesis for the research.
1. Is the adult human hand a unique identifier? We know that the hand shows variation both between individuals, between identical twins and even between the right and left sides of the same individual. There is the possibility that every adult hand is in fact unique, not only because of its inherent genetic variation but because of developmental influences, environmental factors and individual choice to which it is exposed. If the hand is unique, then we do not know the minimum number of anatomical features that would need to be considered to establish that fact, and which features these would be. The only way to reveal the full extent of variation is to be able to study large numbers of images reliably and this is not currently possible due to technical limitations, hence this overriding objective sets the scene for all other relevant questions associated with this project and it relies on the collection of a large ground truth dataset that can be interrogated robustly. Because it would not be possible to physically interact with the large numbers of hands we would need to analyse, the research must focus on the use of images which will be uploaded by the public through citizen science. That images are the source, offers the opportunity for the outcome of the research to be used in the investigation of crimes that involve digital evidence.
2. Can we reliably, repeatedly and automatically extract anatomical information from images of the hand? The biometrics industry has shown that in the presence of the physical hand, we can extract patterns that can be used for identification purposes e.g. fingerprints or vein scanning. However, can we achieve the same levels of reliability from images? Early stage research from our team suggests that this may be possible.
3. Can we still extract that information with the same level of reliability when the hand is in a different dynamic pose or when the image resolution or lighting are not of laboratory standards? In reality, the hand is a mobile structure that can adopt a variety of poses and positions and it is not known how much information is lost by a variation in several degrees of pronation, supination, flexion or extension. Our early research suggests that up to 10% of information might be lost with alteration in lighting or resolution but it is not known if the information lost is ancillary or core to identification capabilities. Therefore, any algorithms developed must be sufficiently robust to be able to adjust to variations in extraneous factors.
4. Image analysis – The current process for assessing human variation from images is inordinately time consuming and therefore restrictive in its ability to address volume issues. Being able to reliably and repeatedly extract anatomical information from images would be a significant development for the identification sciences especially in our digital age. Being able to align what can be seen in 2dimensional images with 3 and 4 dimensional models, would also be a major advancement in the understanding of the representation of detailed human anatomy. The interdisciplinary nature of this research will unquestionably unlock further advances and developments in image analysis that cannot be imagined at this time but its impact on realistic animation for example, may be an avenue for further exploration and exploitation.
5. Biotechnology and bioinformatics – Being able to reconstruct human identity from multimodal biometrics extracted from the hand offers significant opportunities to explore the diversity of the human and to be able to exploit that for the development of software, health assessment or hardware scanners are unlimited. For example, if the hand is proved to be unique, then it may be the only means of identification required and given that it is always carried around by the individual, you could envisage the end of the paper passport or driving licence, no need for written signatures, the end of swipe cards to gain access to high security facilities or the death knell for the lock and key to gain access to our homes or our cars. A multimodal biometric that requires ‘liveness’ indicators e.g. presence of blood pressure, or venous blood would ensure that spoofing is extremely challenging to replicate. Using intelligent filter biometrics would ensure that acquired variation could be adjusted for, e.g. a new tattoo, or scar, and balanced against stable biometrics that are known not to alter with age e.g. fingerprints or vein patterns.