An Investigation of the Reliability of Search Statistics Based on Results from Paired Images
Abstract
For more than 50 years, the Department of Defense and its contractors have been conducting search experiments in the field and with actual and simulated field imagery directed toward development of observer and sensor performance models and psychophysical evaluation of camouflage and signatures of military assets. Model critics have pointed to lack of agreement between model predictions and experimental results, seeking explanations for a perceived lack of correlation. This paper exploits data from a perception experiment to show what can be expected in regard to the consistency of psychophysical quantities such as response time, time for a correct response, average time for a "no target" response, and probability of detection. It examines the difference in consistency between averaged values and raw values from individual subjects. It also examines the difference between average time for a correct response versus the average time for any response (correct or incorrect), as well as the use of average versus median values. The methodology involves the use of paired images. While most of the images in the experiment were shown to the observers only once, a few of the images were shown twice to each observer. The second time the images were shown, they were flipped horizontally (that is, a mirror image was used with a reflection about the vertical axis). Some critics have claimed that, under such circumstances, the observer "learns the image" the first time it is presented. The hypothesized result is that the second time the image is seen, the observer will respond more quickly and more accurately.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2002
- Accession Number
- ADP023104
Entities
People
- James R. Mcmanamey