by James Zjalic
Before CCTV systems were conceived, eye-witness accounts were the only method of identifying a suspect based on their facial features. The pitfalls of this type of identification have been well documented numerous times with regard to the brain’s ability to both forget and modify what a person thinks they witnessed, due to the stress of the situation. It is not surprising therefore that estimations put this method of identification as the greatest factor in wrongful conviction proven by DNA testing .
Facial Comparison, also known as facial mapping, is a field that has seen increased prevalence over the last 20 years due to the proliferation of CCTV cameras in our towns and cities. It has the advantage of not being compromised by stress and the possibility of limiting biases as far as possible. The basic premise is to compare the face of a suspect from a CCTV recording with the face of an individual from a high-quality reference photo (such as a mugshot), to determine the possibility of them being the same person.
There are four methods available to the forensic analyst when performing facial comparisons, listed below.
Holistic, in which all facial features are compared simultaneously. The examiner usually cannot explain the reasoning behind the conclusions reached, so the method is of limited forensic value.
Photo-anthropometric, which is derived from facial anthropometry (the study of facial measurements of individuals using tissue landmarks and bone structure). Photo-anthropometry is the measurement of dimensions and angles of landmarks and other facial features to quantify characteristics and proportions. “Facial-mapping” software makes use of algorithms based on this system.
Superimposition, the alignment of images with one another by creating an overlay using imaging software to make visual comparisons.
Morphological analysis, a systematic method in which a list of facial features is described and the results are compared. A feature list is used for every examination, so the process is replicable. Conclusions are based on the subjective assessment and interpretation of observations.
The Facial Identification Scientific Working Group (FISWG) recommends that morphological analysis is the most reliable and should be used alone or in conjunction with superimposition . Photo-anthropometric analysis is seen on such programs as CSI and Homeland, where computer algorithms will perform complex calculations and output a result to the user in the form of a percentage match or even MATCH in big bold letters.
In reality, forensic images captured by CCTV systems are often are of such low quality due to poor resolution, compression, poor lighting and distance of the suspect from the camera that taking any form of accurate measurement is impossible. The suspect’s pose, camera’s focal length and angle can also add to these quality issues. Increasing the size of an image will not improve the resolution on an individual, as digital images are made up of a finite number of pixels and it will only result in a blocky image in which the pixels can be clearly defined.
One key role of a forensic image analyst is to make the limitations of facial comparison clear to clients, judges and juries whose only familiarity with the process may be from one of the CSI TV shows.
Firstly, a facial comparison can never give 100% certainty regarding a match as it can never be assumed that a face is so individual that nobody in the world has the exact same facial features, especially when those facial features aren’t crystal clear due to the low-quality nature of forensic images.
Secondly, the process is entirely subjective, allowing cognitive biases to sneak in. These biases can be limited through careful operating procedures, such as making observations on the lower quality image first as to not allow confirmation bias to cloud judgment and cause examiners’ brains to turn dead pixels or shadows into a scar that may be present on the higher quality image. Standard operating procedures should always be adhered to, ensuring that subjectivity is minimised .
Finally, there is no agreed upon scale within the field of facial comparison, although there are scales available to the forensic analyst from bodies such as ENFSI and SWIGDE. These scales are usually verbal and are based on the number and quality of positive feature comparisons made. For example, if no unique features (such as unique tattoos or scars) are visible but there are some local features that match (such as hairline or cheekbones) then the result will often be no conclusion but with similarities. It may be reasoned that if there is a result of no conclusion then there would be no reason for the expert to testify as they have arrived at no conclusion. But consider that a police analyst is claiming the suspect in a CCTV recording is the individual they have arrested. If an opposing expert witness puts forward the argument that no conclusion can be drawn, to either exclude or identify the individual, the no conclusion result is, in fact, a conclusion that could show the evidence presented by the police to be inadmissible.
It is more likely that a result of no conclusion will be found than a definite match or exclusion, shown through the phenomenon of “complementary event” taken from statistics. This theorizes that the complementary event (for example finding multiple global features of which none are a unique feature such as a scar) is more likely than a non-complementary event (finding a unique feature such as a scar) .
Requesting reference images of the individual from angles matching that of the CCTV footage allows the process to be optimised and super imposition in support of any findings to be performed. Although certain software options contain tools which assist in changing the perspective of faces, it is not appropriate in this situation as these can change the dimensions of a face by an unknown quantity. The only processing that should be performed on the forensic image is that of enhancement and resizing within constrained proportions. There is software (which will remain unnamed) that instructs the user to flip an image horizontally (mirror image) if the face is turned to the opposite direction of the reference image, allowing their angles to match. This is obviously ill-informed as the human face is not symmetrical and will result in unusable inferences.
As storage gets cheaper and CCTV systems become increasingly present within our lives, so will the number of facial comparisons taking place. People unfamiliar with the digital video recording technology may wonder why the CCTV systems don’t just record at a higher quality resolution, solving the problems of poor resolution and possibly issues with suspects stood a distance from the camera. This would also allow the photo-anthropometric technique to be utilized, removing any subjectivity. The answer is storage.
When recording takes place 24 hours a day at a high resolution such as 1080p, the storage requirements soon become excessive, especially if all recordings are to be backed up indefinitely. There is, therefore, a balance between quality and storage size, but it is expected that the scales will tilt ever more towards higher quality images due to the price of digital storage dropping yearly. The issues with poor lighting and poor angles will still remain, but many cases will become easier and stronger conclusions may be drawn.
In terms of improving the problems with the poor lighting and angles, creating a composite of a suspect’s face from averaging multiple pictures of individuals has been shown to improve the accuracy of photo-anthropometric algorithms . It would also be beneficial if a database was to be created, such that exists within speaker comparison fields, to provide the foundations of a statistical scale to represent the outcomes of an analysis. A combination of progression towards these factors would likely see a movement away from the safety of no conclusion. But for now, we will have to accept that there will be many more no conclusions to come. Luckily, however, facial comparison may be one of the only fields in which the strange occurrence of no conclusion being a conclusion could be the difference between an individual being sent to jail or set free. Innocence Project, “Eyewitness Identification.” Innocence Project, “Eyewitness Identification.”
 FISWG, “Guidelines for Facial Comparison.” 02-Feb-2012. FISWG, “Guidelines for Facial Comparison.” 02-Feb-2012.
 National Crime Agency, CPS, and MET Police, “Forensic Image Comparison and Interpretation Evidence: Guidance for Prosecutors and Investigators.” National Crime Agency, CPS, and MET Police, “Forensic Image Comparison and Interpretation Evidence: Guidance for Prosecutors and Investigators.”
 Graham Upton and Ian Cook, Understanding Statistics. Oxford University Press, 1996. Graham Upton and Ian Cook, Understanding Statistics. Oxford University Press, 1996.
 Dr R. Jenkins and Professor A.M. Burton, “Limitations in Facial Identification: The Evidence.” Vizimetrics, 12-Jan-2008. Dr R. Jenkins and Professor A.M. Burton, “Limitations in Facial Identification: The Evidence.” Vizimetrics, 12-Jan-2008.
About The Author
James Zjalic is a Media Forensics Analyst and partner at Verden Forensics in the UK. Education includes a 1st Class Bachelors Degree in Audio Engineering and an expected Masters Degree in Media Forensics from the National Centre for Media Forensics in Denver, Colorado. Research includes work on image authentication for The Pentagon’s Defense & Advanced Research Project Agency (DARPA) and peer-reviewed publications on subjects including forensic acoustics and audio authentication.
1 thought on “When No Conclusion Can Be A Conclusion”
This is one of the best articles I have come across. Keep up the good work.