The information content of friction ridge impressions as revealed by human experts

Current quantitative analyses of friction ridge impressions likely use only a subset of the information employed by human experts. Translating human expertise into quantitative descriptions of the information available in friction ridge impressions is made difficult by the fact that much of perception takes place below the level of conscious awareness and is difficult to translate into language. The authors approach this problem by collecting eye tracking data from experts to document the regions and features they visit. The authors use data reduction procedures to infer a feature or basis set, and then use this to derive information metrics that provide a complete quantitative description of the information available in friction ridge impressions given a particular feature set and metric. The range of potential information metrics is large, and the authors propose to use a recursive testing methodology that uses data reduction procedure to obtain an initial feature set, which will then be used to generate candidate information metrics. These metrics will then be tested using a combination of noise masking and feature elimination procedures depending on the metric used. The researchers will explore a range of metrics as well as combinations of information sources to reflect the fact that experts may use a wide variety of different types of features. The final test will come when they use the metrics to predict the eye fixations of experts on new sets of fingerprints, which will demonstrate that the experts and the researchers' quantitative analysis both rely on similar sources of information as a way to validate their analyses. The specification of a metric also allows the researchers to determine which features are most diagnostic and point out when experts are using sub-optimal search strategies. Finally, the researchers will use the data to identify different search patterns that are associated with the different steps involved in the ACE-V methodology, and reveal behaviors that lead to errors. ca/ncf