Show, don't tell: Non-verbal eyewitness testimony based on non-artistic face sketches

Abstract

In this text we focus on the problem of eyewitness face sketch recognition, which over 30 years of psychological studies have showed that currently used eyewitness testimony procedures (ETPs) are inaccurate and highly unreliable. Moreover, current automatic face sketch recognition methods (FSRs) cannot be applied for recognizing forensic sketches. We here propose to use non-artistic sketches directly drawn by the eyewitness (Main Sketches), as the medium to retrieve eyewitness' mental image of the target face. In our FSR, we firstly estimate the sketching bias for each eyewitness and debias the Main Sketch, and then match this estimation to the photo database, using a weighted dynamic point correspondence, inspired by psychological suggestions for face perception in humans. We test our proposed method on a total of 860 sketches, drawn by 86 human participants, and showed that we can achieve significantly better performances in contrast to previous methods.