A perturbation method for evaluating background subtraction algorithms

TitleA perturbation method for evaluating background subtraction algorithms
Publication TypeJournal Articles
Year of Publication2003
AuthorsChalidabhongse TH, Kim K, Harwood D, Davis LS
JournalJoint IEEE International Workshop on Visual Surveillance and Performance Evaluation of Tracking and Surveillance (VS-PETS)
Date Published2003///

We introduce a performance evaluation methodology calledPerturbation Detection Rate (PDR) analysis, for measuring
performance of background subtraction (BGS) algorithms.
It has some advantages over the commonly used Receiver
Operation Characteristics (ROC) analysis. Specifically, it
does not require foreground targets or knowledge of fore-
ground distributions. It measures the sensitivity of a BGS
algorithm in detecting low contrast targets against back-
ground as a function of contrast, also depending on how
well the model captures mixed (moving) background events.
We compare four algorithms having similarities and differ-
ences. Three are in [2, 3, 5] while the fourth is recently de-
veloped, called Codebook BGS. The latter algorithm quan-
tizes sample background values at each pixel into code-
books which represent a compressed form of background
model for a long image sequence.