視訊剖析與索引之研究
Other Title
A Study on Video Parsing and Indexing
Date Issued
1999
Date
1999
Author(s)
DOI
882213E002060
Abstract
Many indexing methods have been
proposed. However, most of them are
built on the basis of distance between
feature vectors. The indexing methods
based on the distance can not handle the
problem of scalar shifts. So, we propose
a new index method, called ConeTree,
which is suitable for nearest neighbor
query based on the angle between two
features vectors. Thus, our ConeTree can
deal with the problem of scalar shifts.
Instead of partitioning the data space into
rectangular or spherical regions, the
Cone-tree partitions the data space into
cone-shape regions. The smaller the
distances among points are, the closer
the points are clustered in the SR-tree.
Similarly, the smaller the angles among
feature vectors, the closer the feature
vectors are clustered in the Cone-tree.
Our Cone-tree is a balanced tree whose
height is bounded by O(logc(N)), where
N is the number of feature vectors and c
is the maximum number of feature
vectors stored in a node. In the case of
searching based on the angle between
two features vectors, the results of our
experiments demonstrate that a Cone-tree outperforms an SR-tree in most of
cases.
Subjects
Video data
Video database
Video frame
Shot
Scene
Index
Index tree
Publisher
臺北市：國立臺灣大學資訊管理學系暨研究所
Type
other
File(s)![Thumbnail Image]()
Loading...
Name
882213E002060.pdf
Size
20.74 KB
Format
Adobe PDF
Checksum
(MD5):7a375098d85f2dbb84a56f8f089808c1