Computer assisted pattern recognition

Abstract

Recent advances in computer technology, specifically artificial intelligence, have provided the human factors community with a new opportunity; to devise general rules for integrating effectively real-time computer pattern recognizers (CPRs) with the slower but more discriminating human pattern recognition capability. Efficient CPRs are fast and comprehensive pattern filters, which reduce human task workload from arduous free search or gross information extraction to simple binary decisions, by essentially providing a much smaller "candidate" pool of pre-filtered information. CPRs are characterized by their ability to discriminate signal from noise (a probability function). CPRs are more likely to err by selecting false candidates while humans select very few false candidates. Computer pattern recognition is being developed for a variety of applications, many involving complex imagery interpretation. These include many medical applications, from interpreting standard X-rays to magnetic resonance imaging. Several quality control tasks in manufacturing could be enhanced by CPR technology, such as finding near microscopic faults in metal composites. There is also increased military interest in CPR technology. Sensor images, such as FLIR (forward looking infrared), radar, or image-intensified television, are often used to detect

military targets and threats. Human search time of large amounts of sensor imagery, and consequently vulnerability to threats, could be greatly reduced by effective use of CPRs, reducing the amount of information to be scanned. System developers, specifications writers, and potential customers are disadvantaged when applying CPR technology to the real world because they cannot predict the magnitude of the total system's (human plus CPR) performance enhancement prior to testing the finished prototype. Current research in the areas of decision aiding, cuing, and free search exposes the technical community to the myriad of man-machine interface issues, but the usable knowledge tends to be task specific. The need for a generalizable and predictive framework for effective human-system interaction clearly exists. This research advances a methodology for determining the performance effectiveness of the total human-CPR system, given particular task and CPR characteristics. The paradigm for the predictive model is the Receiver Operating Characteristic, or ROC, taken from Signal Detection Theory (SDT). The relatively high workload task is to discriminate the presence of target objects in a matrix of similar nontarget objects within a variable time interval. The selected task is representative of those with which CPR technology is currently being applied. A graphics computer program was developed to present target object matrices on - the monitor. An emulated CPR of varying diagnostic ability

was used to search the object array and to ''preview" and identify candidate patterns as likely targets. The subject then viewed the object matrix and either verbally agreed or disagreed with the CPR's target nominations. Any available spare time was used to search for additional target objects which the CPR may have missed. One condition involved presentations of the same types of matrices, but without any CPR preview. This essentially became a free search task and served as a comparison baseline. Hits and false alarms were recorded. For each search time period, a series of ROC curves was generated for a variety of payoff conditions. A model can be created when system developers are interested in how much performance improvement can be expected to occur when a CPR of a given diagnostic ability is added to a particular type of task. For shorter search times, CPRs in general enhanced task performance over the unassisted condition. Given a longer search time, the CPR's assistance diminished in its overall performance enhancement effect. Moreover, for the short time period, the best of the diagnostic CPRs showed the greatest improvement. False alarm rates were very low; they were lowest when the 90% CPR was used. Payoff influenced hit rates significantly, but hit rates were most significantly affected by CPR diagnostic ability. From this model, one can predict actual system performance (human plus CPR) given the: a) baseline task performance of the unaided human, b) diagnostic ability of the CPR, c) payoff characteristic of the signal detection task, and d) time available to complete the task. Included in the research effort is an application of the model and its design implications relative to the development of military aided target recognizers (ATRs).

Notes

This item is only available in print in the UCF Libraries. If this is your thesis or dissertation, you can help us make it available online for use by researchers around the world by STARS for more information.

Graduation Date

1992

Semester

Spring

Advisor

Gilson, Richard D.

Degree

Doctor of Philosophy (Ph.D.)

College

College of Arts and Sciences

Department

Psychology

Format

PDF

Pages

158 p.

Language

English

Length of Campus-only Access

None

Access Status

Doctoral Dissertation (Open Access)

Identifier

DP0029864

Subjects

Arts and Sciences -- Dissertations, Academic; Dissertations, Academic -- Arts and Sciences

This document is currently not available here.

Share

COinS
 

Accessibility Statement

This item was created or digitized prior to April 24, 2027, or is a reproduction of legacy media created before that date. It is preserved in its original, unmodified state specifically for research, reference, or historical recordkeeping. In accordance with the ADA Title II Final Rule, the University Libraries provides accessible versions of archival materials upon request. To request an accommodation for this item, please submit an accessibility request form.