Forensic Facial Identification using the Fringe P3 brain wave response (EEG-FIT)

从目击者证词(俗称照相或伪装)中制作犯罪嫌疑人的面部复合体是世界各地协助刑事调查的主要工具。这个项目提出了一种激进的、新的方式来获得这类图像，提高了准确性和速度。目前，犯罪目击者在训练有素的警察操作员的指导下生成合成图像，这一过程通常涉及广泛的语言互动。因此，这一过程对人力资源要求很高(需要1-4个小时)，而且准确度相当低。项目合作伙伴Visionmeeter和肯特大学的霍华德·鲍曼教授旨在通过结合他们的专业知识并取得必要的进展来克服当前方法的局限性。其核心思想是向犯罪目击者展示计算机生成的图像的快速序列，这些图像经过计算后与嫌疑人相似。脑电(EEG)被用来捕捉增强的脑波反应，这种反应是对与罪犯相似的刺激所观察到的。这种方法(称为条纹P3方法)已经由肯特大学的鲍曼广泛开发。VisionMetric专门研究机器学习和应用于人脸的人工智能方法。他们建议开发快速和健壮的迭代过程，从P3信号流中生成越来越相关的面部图像，并将这些图像呈现给目击者。它直接利用目击者的正常脑波活动，避免了目击者和操作员之间广泛互动的需要。这样做的目的是为了证明，可以在现有方法所需时间的一小部分内生成更准确的嫌疑人图像。

"The production of facial composites of criminal suspects from eyewitness testimony (commonly known as PhotoFITs or EFITs) is a staple tool used to assist criminal investigations throughout the world. This project proposes a radical, new way of obtaining such images with increased accuracy and speed.At present, eyewitnesses to crime produce composite images under the guidance of a trained police operator and the process generally involves extensive verbal interaction. The process is thus demanding on human resources (taking 1 - 4 hours) and accuracy is quite low.The project partners Visionmetric and Prof Howard Bowman, University of Kent aim to overcome the limitations of current methods by combining their expertise and making the necessary advances. The central idea is to present eyewitnesses to a crime a rapid sequence of computer-generated images calculated to resemble the suspect. EEG (electroencephalography) is used to capture an enhanced brain wave response that is observed in response to stimuli which bear resemblance to the criminal offender. This method (known as the fringe P3 method) has been extensively developed by Bowman at University of Kent. Visionmetric specialise in machine learning and AI methods applied to the human face. They propose to develop fast and robust iterative processes for generating increasingly relevant facial images from the P3 signal stream and presenting these to the witness.This approach is very fast. It directly exploits the normal brainwave activity of the eyewitness and avoids the need for extensive interaction between the witness and operator. The aim will be to show that more accurate images of a suspect can be produced and in a small fraction of the time that is required using existing methods."