Grant: Discovery Grants Program - Individual

Eyewitness identification is often highly regarded by the legal system, yet it is not always accurate. An identification may be compromised if the words used to describe the target face impair the witness' memory of that face (verbal overshadowing) or if the witness sees so many photos that he or she may identify a photo as the target face when it is actually similar to one of the early photos viewed (inaccurate source attribution). Faces of a race different than the witness may cause additional difficulties (cross-race effect).In order to improve the accuracy of eyewitness identification, these sources of inaccuracies must be mitigated. Verbal overshadowing may be addressed by using only non-verbal means to ask for recognition, rather than recall, of the target face. Inaccurate source attribution may be ameliorated by reducing the number of candidate photos presented to a witness. In order to meaningfully achieve this reduction, it is necessary to understand how a particular witness judges similarity between photos so that a computer-based presentation can be personalized accordingly. My research to date indicates that people employ one of at least two different strategies when judging facial similarities. I seek to democratize accurate identification of faces by: understanding the different strategies that people may use to judge facial similarity; developing a concise means of testing which strategy a witness is likely to apply; developing a computer interface that makes use of a witness' identified strategy to allow accurate identification of the target face, while viewing as few candidate photos as possible; and developing an extensible system to manage and store candidate photos for use with the interface. My research to date has used a set of 356 photos (178 Caucasian and 178 First Nations to allow for examination of cross-race effect), which participants have sorted based on how they each judge similarity. This work will contribute to eyewitness identification by providing better tools for witnesses and by providing a basis for understanding which faces may be more or less easy to identify. The same techniques can be used to study interaction with fractal images, which has applications for democratization of multimedia composition.