News

Facial compositing software will benefit criminal justice process: Academics

  • Note high rate of failure in relying on hand-drawn composite sketches

By Ruwan Laknath Jayakody

An automated, image processing-based computer software solution that also has two dimensional (2D) facial feature templates will enhance the manual facial composite procedure, which will be of benefit when it comes to providing services to the criminal justice process, such as to law enforcement officers and legal counsel, and to thereby ensure that victimised persons have quick access to justice.

This was noted in a research article on “Forensic art: 2D facial composite through image processing techniques in Sri Lanka”, which was authored by L. Sivaneasharajah, M.A.S. Perera, P.B. Jayasekara, D.D. Karunaratne, K.D. Sandaruwan, and R.N. Rajapakse (all attached to the Colombo University’s School of Computing) and J. Perera (attached to the Colombo University’s Medical Faculty’s Forensic Medicine and Toxicology Department) and published in the Sri Lanka Journal of Forensic Medicine, Science, and Law 7 (2) in February 2017.

A facial composite constitutes a graphical representation of a human face according to, as noted by Sivaneasharajah et al., the description of eyewitnesses.

In Sri Lanka, in order to obtain a facial composite, forensic art involving the traditional procedure of manual hand drawing is utilised for the purpose of identifying suspects. However, as per a personal communication made during an interview to Sivaneasharajah et al. by the Criminal Records Division of the Police on the “Manual process of suspect identification through facial composite in Sri Lanka”, the failure rate of suspect identification through the manual facial composite sketch was reportedly quite high (92.86% in 2014).

Therefore, Sivaneasharajah et al., pointed out that this higher failure rate means that there is a need for a solution concerning the facial composite as the inability to identify perpetrators of crime leaves the criminal justice system in a precarious position.

In “Facial index based 2D facial composite process for forensic investigation in Sri Lanka”, P.B. Jayasekara, L. Sivaneasharajah, M.A.S. Perera, J. Perera, D.D. Karunaratne, K.D. Sandaruwan, and R.N. Rajapakse introduced an automated, information technology (IT) – specifically image processing-based, computer software solution which has 2D facial feature templates that incorporate medically defined indices and aesthetic aspects, along with statistical analysis techniques and related aspects.

Although C. Frowd, D. McQuiston-Surrett, S. Anandaciva, C. Ireland, and P. Hancock’s “An evaluation of the US systems for facial composite production” notes that there are numerous such international software that are available for the purpose, these software are not best suited to be adopted to the Sri Lankan context as the facial feature templates which are available in these software are not relevant for the local population (per the Police Criminal Records Division).

Hence, Sivaneasharajah et al. explained that as the first step, facial feature templates need to be constructed by analysing a dataset from the local population.

In this regard, “Facial muscle anatomy-based approach for forensic facial reconstruction in Sri Lanka” by R. Rajapakse, A. Madugalla, I. Amarasinghe, V. Padmathilake, A. Dharmaratne, D. Sandaruwan, and M. Vidanapathirana found that the ability to assess the facial feature appearance has been achieved by anthropometric proportion indices, which has been used in a number of forensic practices. Therefore, Sivaneasharajah et al. noted that when constructing the facial feature templates, two major parameters known as most occurring indices and commonly available facial feature shapes have been used.

Furthermore, so as to find the relative measurements for these parameters, two sub-researches on facial anthropometric indices and shape classification were carried out targeting the local population. In the first phase of the research, these sub-researches were conducted with 140 male and female undergraduate students, between the ages of 20 and 25 years, in order to find out the anthropometric proportion indices measurements and the shape classification for the eyes, nose, face, upper vermilion (the vermilion border is the normally sharp demarcation between the lip and the adjacent normal skin), and lower vermilion for Sri Lankans.

Moreover, the entire data set was categorised based on the weight categories according body mass index values used to screen different weight categories. The index measurements were measured with the use of Face SDK library (a multi-platform, face recognition, identification, and facial feature detection solution, which is comparatively reliable and accurate when compared with manual measuring techniques) while domain expert’s knowledge from the aesthetic and forensic art field were utilised in identifying the facial feature shapes.

Thereafter, by incorporating these parameters (weight categories; index measurement of the facial features; commonly available shapes of facial features; 2D facial feature templates; facial feature template creation process by incorporating anthropometric indices and commonly available shapes), 2D facial feature templates for the face, eyes, nose, and mouth were created targeting Sri Lankan people, and afterwards these templates were transferred to a system which is used for the purpose of composite construction.

This system also enables the filtering of the most possible feature templates based on the eyewitness description given with the age, height, weight, and other criteria. This research study has identified the template positioning of each facial feature by analysing the distance between the facial features and incorporating the standard proportions suitable for the Sri Lankan context, as per Y. Jefferson’s “Skeletal types: Key to unravelling the mystery of facial beauty and its biological significance”. In the end, the face visualisation process will perform iterations until the eyewitness is satisfied with the ultimate composite image.

Jayasekara et al. evaluated this solution with face pool (70.19% accurate) and anthropometric (an average of 84% out of the population that gave a response with regard to the matching of the faces) index evaluation methods and techniques.

“This shows the significance of the accuracy level in composite images, when it uses the anthropometric indices measurements. Hence, this fully implemented computerised solution will enhance the current manual facial composite procedure, and eventually this solution will benefit relevant personnel who provide services to the criminal justice process such as law enforcement officers and legal counsel in order to ensure that victimised persons have quick access to justice,” Sivaneasharajah et al. concluded.