|
Monday David Lowe Object Recognition using Invariant Local
Features Human vision is so powerful that we
seldom give a second thought to our ability to immediately identify the objects
in our surroundings. However, the problem of object recognition has proved very
challenging for computer vision. A major source of the difficulty is the large
range of variations in appearance that may occur, due to factors such as
changes in 3D viewpoint, varying illumination, partial visibility, and
background clutter. Fortunately, there has been rapid progress on this problem
within the past few years through the use of an approach known as invariant
local feature matching. Thousands of these local features can be extracted from
an image to describe small overlapping regions, and each feature is designed to
be invariant to a range of image transformations, such as changes in scale,
orientation, brightness, and local deformations. Furthermore, the features are
designed to be very distinctive, so that a single feature can be used to select
a correct match from a large database. Fast methods for nearest-neighbor access
in high-dimensional spaces allow large databases of features to be matched in
real time. This talk will present an overview of the invariant feature
approach, as well as some recent applications such as location recognition and
automated stitching of digital images into panoramas. Biography: David Lowe is a professor of Computer
Science at the |
_______________________________________________ Colloquium mailing list [email protected] https://secure.engr.oregonstate.edu/mailman/listinfo/colloquium
