[talks] C Barnes preFPO

Melissa Lawson mml at CS.Princeton.EDU
Wed Oct 6 10:55:23 EDT 2010

Connelly Barnes will present his preFPO on Wednesday October 13 at 3PM in Room 302 (note
The members of his committee are:  Adam Finkelstein, advisor; Szymon Rusinkiewicz and Dan 
Goldman (Adobe), readers; Tom Funkhouser and Bob Sedgewick, nonreaders.  Everyone is
to attend his talk.  His abstract follows below.

PatchMatch: A Fast Randomized Matching Algorithm with Applications to
Images, Videos, and 3D Volumes


This thesis presents a novel fast randomized matching algorithm for
finding correspondences between small local regions of signals. We
also explore a wide variety of applications of this new fast
randomized matching technique.

The core matching algorithm, which we call PatchMatch, can find
similar regions or "patches" of an image one to two orders of
magnitude faster than previous techniques. The algorithm requires only
very loose assumptions that neighboring correspondences tend to be
similar or "coherent" in order to quickly converge to an approximate
solution. Our subsequent research has shown that the matching
algorithm is quite generic, and can be extended to work on video
summarization, computer vision problems, collections of images
(ongoing work), and 3D volumes (ongoing work). Our fully generalized
algorithm works on a variety of 2D and 3D signals in quite disparate
application domains, gaining speed-ups over alternative techniques in
a number of areas.

We have explored many applications of this matching algorithm. In
computer graphics, we have explored removing unwanted objects from
images, seamlessly moving objects in images, changing image aspect
ratios, and video summarization. In computer vision we have explored
denoising images, object detection, detecting image forgeries, and
detecting symmetries. In our ongoing work we plan to also apply our
algorithm to collections of images, and producing hybrids from a set
of input exemplar meshes. We finally will discuss how the statistics
of the inputs relate to convergence of our algorithm, the limitations
of our algorithm, and areas for future research.

More information about the talks mailing list