Human fusion of image and numeric information in machine-aided target recognition

An operator's ability to use both numeric and image data to detect targets is a critical issue for machine-aided target recognition. This paper uses classic Bayesian and quasi-Bayesian models to estimate target-detection rates based on the combination of image and numeric evidence. The quasi-Bayesian model, fitted to actual detection-rate data, indicates that operators do not appropriately adjust their reliance on numeric data as the quality of that data changes relative to the quality of image data. Results show that operators decrease rather than increase their reliance on numeric data as its quality increases relative to the quality of images. The results suggest that operators working with an automated target recognition system may have difficulty in assessing the value of the system's numeric judgments in comparison with their own judgments based on images.<
>