Using the smooth Receiver Operating Curve (ROC) method for evaluation and decision making in biometric systems

PDF

Authors
  1. Gorodnichy, D.
  2. Granger, E.
  3. Matwin, S.
Corporate Authors
Defence Research and Development Canada, Centre for Security Science, Ottawa ON (CAN);Canada Border Services Agency, Science and Engineering Directorate, Ottawa ON (CAN);OTTAWA UNIV, OTTAWA ON (CAN) SCHOOL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCE
Abstract
This report presents the use of the smooth ROC method for evaluation of biometric systems, which can be used in developing decision making rules for face recognition triaging. From a performance evaluation perspective, the problem can be decomposed into two subproblems: the first deals with detecting the presence or absence of a specified person of interest (POI) in a given video frame, and the second measures the strength of matching between the POI image and the face found in the video frame. The former can be viewed as a binary decision of detecting (or the lack there of), and the latter is based on a score that depicts the strength of matching. Ideally, the objective of the system is to measure the agreement between higher matching scores with the presence of POI in the video frame based on the assumption that a higher matching score corresponds to a higher likelihood of the POI being present in the frame. The accumulation of this performance information across the stream of video frames will yield information required for performance assessment analysis of the system as a whole.
Keywords
biometrics;surveillance;receiver operating curve (ROC)
Report Number
DRDC-RDDC-2014-C169 — Contract Report
Date of publication
01 Jul 2014
Number of Pages
17
DSTKIM No
CA039500
CANDIS No
800406
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: