Campus Event Calendar
Event Entry
What and Who
Background estimation from non-time sequence images
Miguel Granados
IMPRS
Masterseminar/Talk
AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, RG2
Public Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Tuesday, 4 December 2007
13:15
45 Minutes
E1 4
024
Saarbrücken
Abstract
Given a set of non-time sequential photographs, which are taken using
the same view point, camera settings, and
similar illumination conditions, we address the problem of estimating
the photographed scene's background. One
example application would be to capture a clean, unoccluded shot of a
crowded public place, relevant in artistic
photography or for VR/AR applications. We approach the problem by
first defining the scene's background as a
labeling over the set of input images, assigning each labeling a cost,
and finally minimizing the associated cost
function. The labeling cost penalizes deviations from the following
two model assumptions: background objects
are more likely to appear across the input images, and background
objects are stationary. We approximate object
likelihood using a per-pixel density estimation, and object
stationariness using a motion boundary consistency
term. To correctly compute the likelihood we use a decorrelated color
space. The cost function is minimized using
the expansion move algorithm, and the final result computed using
Poisson blending. Our contribution is the
definition of an automatic method for background estimation based on
an significantly improved energy function,
which outperforms previous results.
the same view point, camera settings, and
similar illumination conditions, we address the problem of estimating
the photographed scene's background. One
example application would be to capture a clean, unoccluded shot of a
crowded public place, relevant in artistic
photography or for VR/AR applications. We approach the problem by
first defining the scene's background as a
labeling over the set of input images, assigning each labeling a cost,
and finally minimizing the associated cost
function. The labeling cost penalizes deviations from the following
two model assumptions: background objects
are more likely to appear across the input images, and background
objects are stationary. We approximate object
likelihood using a per-pixel density estimation, and object
stationariness using a motion boundary consistency
term. To correctly compute the likelihood we use a decorrelated color
space. The cost function is minimized using
the expansion move algorithm, and the final result computed using
Poisson blending. Our contribution is the
definition of an automatic method for background estimation based on
an significantly improved energy function,
which outperforms previous results.
Contact
IMPRS
9325-225
--email hidden
passcode not visible
logged in users only
Andrea Primm, 11/29/2007 09:49
Andrea Primm, 10/25/2007 11:47
Andrea Primm, 10/09/2007 11:31 -- Created document.
Andrea Primm, 10/25/2007 11:47
Andrea Primm, 10/09/2007 11:31 -- Created document.