New Update Forensic Facial Regeneration and Recognition Software

PhotoFIT and Identikit tools have been helping the police identify suspects from eye witness statements for years, today however; the computerized versions of these tools rely on reconstructing a face from half-remembered and often poorly described features.
According to Stuart Gibson of the University of Kent, Canterbury, and colleagues at the Open University, Milton Keynes, the effectiveness of this feature-based approach is fundamentally limited by the ability of the witness to recall and describe the suspect. The older version of this process requires the witness to sit with a modern day sketch artist, and give a detailed description of every feature starting with head shape and moving all the way through to such details as eye shape and positioning. This old system relies heavily on the witness’s ability to recall every feature perfectly completely from memory; here in lies the problem. Eye witness testimonies are among the most unreliable and the courts more often than not – when an arrest is made based off of a computer re-generation of the perp. is often put through the mill in terms of accuracy. However, in recent research Gibson and co-workers developed a new approach that allowed the whole face to be recreated in a much more intuitive way. The EFIT-V software developed from the team's research is now being used by a number of police services in the UK.
"Unlike traditional feature based methods, the approach described here utilizes global, whole face, facial characteristics and allows a witness to produce plausible, photo-realistic face images in an intuitive way," says Gibson and colleagues.
With EFIT-V, the operator asks the witness some very simple questions relating to the age, sex, face-shape and hairstyle of the suspect the police are hoping to identify, explains Gibson. This information is used to initializes the EFIT-V system. In the steps that follow the witness is shown a set of computer-generated faces from which they must select the face that represents the best likeness to the suspect. Variants of the selected face are then generated and these comprise a new set of images from which another selection is made.
A good likeness to the suspect can be achieved simply by repeating this process a number of times. This new approach, which effectively ‘evolves’ a facial likeness, latches on to a person's ability to instantly recognize a face when they see it again which is known to be superior to the human capability for recalling in detail a face that was only seen fleetingly, the researchers say.
The witness is not required to provide a detailed verbal description of the face, so a lengthy and potentially tiring interview can be avoided. This new process also eliminates the long and tedious process which often causes poor, even blatantly wrong choices during the description process because in order for the old images to be created a choice must be made as to each feature so if the witness could not recall their eye color, a random one would be chosen.
Gibson and colleagues point out that controlled trials supported by the UK Home Office demonstrated that the system could double the useful intelligence gained compared with conventional approaches. Twelve UK police services are now using the system successfully.