Abstract
Despite popular misconceptions, crime-scene fingerprint identification is not an automated process. Instead, human examiners must visually match latent fingerprints collected from crime scenes to references from potential suspects. Past research has indicated an effect of source finger type on accuracy when identifying fingerprints. However, it is unknown whether this effect is due to differences in the information each digit provides, or how that information is processed. Therefore we conducted an experiment to investigate the influence of source finger type, image similarity, and level of available stimulus information on response accuracy in a same-different task that required naive subjects to indicate whether pairs of latent and reference fingerprints matched. Latent fingerprints were processed using principle component analysis to vary the amount of stimulus information, that is, the percentage of variance explained in each image when compared to the fingerprint eigenspace. To determine the effect of image similarity, non-matching reference prints were of either low or high cosine similarity to latents. Our findings replicated those of previous studies, demonstrating that identification accuracy depends on source finger type and image similarity, and that no digit/similarity interaction was present. Higher accuracy was associated with the thumb, middle digit, and dissimilar images, while that for the little finger was lower. Signal detection analyses revealed a positive linear relationship between identification accuracy and the percentage of variance accounted for in latent prints. We also found an overall negative relationship between accuracy and response time, and that the accuracy/RT relation depends on the source digit and amount of stimulus information. Currently we are conducting experiments examining the effects of image quality on response accuracy. However, preliminary evidence suggests that the effect of digit on accuracy is due to differences in how fingerprints are processed, rather than the information each digit provides.
Meeting abstract presented at VSS 2016