INTRODUCTION

Forensic art is a technique used by law enforcement in order to identify a suspect towards apprehension or conviction of wanted persons and composite art is one branch. A “facial composite” is a graphical representation of a human face according to eyewitnesses' description. The failure rate of suspect identification through manual facial composite sketch was reportedly quite high in 2014 in Sri Lanka¹. In addition, international software available for face identification may not be the best approach for Sri Lankan population.

OBJECTIVE

To develop an automated, image processing based computer software solution using 2D facial feature templates for suspect identification in Sri Lanka.

STUDY DESIGN

Facial feature templates were constructed by analyzing a dataset from Sri Lankan young adult population focusing on most occurring facial indices and commonly available facial feature shapes. Anthropometric proportion indices measurements and shape classification for eyes, nose, face, upper vermilion and lower vermilion of 140 undergraduates of both sexes between 20-25 age were identified and categorized according to BMI (Body Mass Index). Face Software Development Kit library (Face SDK library) was used for index measurements and domain expert knowledge from aesthetic and forensic art fields were utilized in identifying facial feature shapes. These parameters were incorporated to construct 2D facial feature templates for major facial feature. Then these templates were transferred into the system to be used for composite construction purpose. Finally face visualization process will perform iteratively until the eyewitness is satisfied with the ultimate image. Evaluation of this solution with face pool technique and anthropometric index evaluation techniques obtained 70.19% accuracy and average of 84% respectively.