A Perceptual Scaling Approach to Eyewitness Identification

长期以来，目击者辨认在刑事调查和起诉中发挥着重要作用，在刑事司法程序中，辨认罪犯是至关重要的。基于对感知和记忆的理解的进步，本项目将研究“填充者”在队列中的作用。填充者指的是一组已知无辜的人，他们用来挑战目击证人的指认。该研究将开发一种基于人机面部相似性测量相结合的填充物选择新方法。该项目建立在设计和实施目击者识别队列的新范例之上。利用知觉尺度，在感觉心理学中建立，这种方法将用于估计目击者记忆的强度，从而产生大量的面部图像库，其中面部相似性的程度是量化的。这项研究将有助于减少目击者对决策偏见的敏感性，并将提供对面孔感知相似性的定量测量。这些措施与机器学习算法相结合，将用于创建模型证人，该模型证人是一种复制人类观察者的相似性判断并预测新面孔感知相似性的算法。该项目将产生一个客观的系统来选择填充物，以优化目击者的公平性和性能。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Eyewitness identification has long played an important role in criminal investigations and prosecutions, where identifying the culprit is critical to the criminal justice process. Based on advances in understanding of perception and memory, this project will examine the role of “fillers” in lineups. Fillers are the individuals in a lineup known to be innocent who serve to challenge eyewitness recognition. The research will develop a novel method for filler selection based on a combination of human and machine measures of facial similarity.The project builds upon a new paradigm for the design and conduct of lineups for eyewitness identification. Drawing upon perceptual scaling, well-established in sensory psychology, this approach will be used to estimate the strength of eyewitness memories, and thus to generate large libraries of facial images in which the degree of facial similarity is quantified. This research will help to reduce susceptibility of the eyewitness to decision bias and will provide quantitative measures of the perceived similarity of faces. These measures, in combination with a machine learning algorithm, will be used to create a model witness, which is an algorithm that reproduces similarity judgements of human observers and predicts perceptual similarity for new faces. The project will yield an objective system for selecting fillers in order to optimize eyewitness fairness and performance.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.