BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Microsoft Research Cambridge\, public talks
SUMMARY:Efficient Data Structures for Nonlinear Video Proc
essing - Jiawen Chen\, MIT's Computer Science and
Artificial Intelligence Laboratory
DTSTART;TZID=Europe/London:20110411T140000
DTEND;TZID=Europe/London:20110411T150000
UID:TALK30741AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/30741
DESCRIPTION:Nonlinear techniques are used extensively in image
and video processing with applications ranging fr
om low level kernels such as denoising and detail
enhancement to higher level operations such as obj
ect manipulation and special effects. In this talk
\, we will describe two computationally efficient
data structures which dramatically simplify and ac
celerate a variety of algorithms for video process
ing.\n\nOur first data structure is the bilateral
grid\, an image representation that explicitly acc
ounts for intensity edges. By interpreting bright
ness differences as Euclidean distances\, the bila
teral grid naturally encodes the notion of edge-aw
areness into filters defined on it. Smooth functi
ons defined on the bilateral grid are piecewise-sm
ooth in image space. Within this framework\, we d
erive efficient reinterpretations of a number of n
onlinear filters commonly used in computational ph
otography as operations on the bilateral grid\, in
cluding the bilateral filter\, edge-aware scattere
d data interpolation\, and local histogram equaliz
ation. We also show how these techniques can be e
asily parallelized onto modern graphics hardware f
or real-time processing of high definition video.\
n\nThe second data structure we describe is the vi
deo mesh\, designed as a flexible central data str
ucture for general-purpose nonlinear video editing
workflows.\nIt represents objects in a video sequ
ence as 2.5D "paper cutouts" and allows interactiv
e editing of moving objects and modeling of depth\
, which enables 3D effects and post-exposure camer
a control. In our representation\, motion and dep
th are sparsely encoded by a set of points tracked
over time. The video mesh is a triangulation ove
r this point set and per-pixel information is obta
ined by interpolation. To handle occlusions and d
etailed object boundaries\, we rely on the user to
rotoscope the scene at a sparse set of frames usi
ng spline curves.\nWe introduce an algorithm to ro
bustly and automatically cut the mesh into local l
ayers with proper occlusion topology\, and propaga
te the splines to the remaining frames. Object bo
undaries are refined with per-pixel alpha mattes.\
n\nAt its core\, the video mesh is a collection of
texture-mapped triangles\, which we can edit and
render interactively using graphics hardware. We
demonstrate the effectiveness of our representatio
n with special effects such as 3D viewpoint change
s\, object insertion\, depth-of-field manipulation
\, and 2D to 3D video conversion.\n
LOCATION:Small lecture theatre\, Microsoft Research Ltd\, 7
J J Thomson Avenue (Off Madingley Road)\, Cambrid
ge
CONTACT:Microsoft Research Cambridge Talks Admins
END:VEVENT
END:VCALENDAR