ITEE seminar: Dr Heng Tao Shen, 10.00AM, Mon 27 Oct 2003

Efficient Database Support for WWW Image Retrieval

Speaker: Dr Heng Tao Shen, National University of Singapore

When: 10.00AM, Monday 27 Oct 2003

Venue: 78-420

Host: Professor Maria Orlowska

Abstract:

  WWW is exploring and shaping the current research direction.  To
  enhance the WWW page content, images are increasingly being embedded
  in web pages. Such pages over the WWW essentially provide a rich and
  interesting source of image collection from which users can
  query. Uniquely, WWW images carry both high-level feature - text,
  and low-level features - color, shape, and texture. Typically, each
  feature is represented as a high-dimensional point (or feature
  vector).

  Unfortunately, most WWW image search engines fail to exploit image
  semantics and give rise to low precision. On the other hand,
  existing indexing techniques fail to provide more efficient
  retrieval than sequential scan as the dimensionality of image
  features reaches high due to the well-known 'dimensionality
  curse'. Moreover, the problem of indexing multiple image features is
  too hard to have been addressed. To build an image retrieval system,
  both effectiveness and efficiency have to be considered.

  In this talk, we focus on two novel indexing methods which provide
  strong efficiency support for the rertieval.

  1): One well known approach to overcoming 'dimensionality curse' is
    to reduce the dimensionality of the original dataset before
    constructing the index. We present an adaptive Multi-level
    Mahalanobis-based Dimensionality Reduction (MMDR) technique for
    high-dimensional indexing. Our MMDR technique has four notable
    features compared to existing methods. First, it discovers
    elliptical clusters for more effective dimensionality reduction by
    using only the low-dimensional subspaces. Second, data points in
    the different axis systems are indexed using a single B+-tree.
    Third, our technique is highly scalable in terms of data size and
    dimension. Finally, it is also dynamic and adaptive to insertions.

  2): To futher support hyper-dimensional databases which contain
    hundreds or even thousands of dimensions.  We introduce a novel
    methodology called Local Digital Coding (LDC). LDC extracts a
    simple bitmap representation called Digital Code(DC) for each
    point in the database.  Pruning during KNN search is performed by
    dynamically selecting only a subset of the bits from the DC based
    on which subsequent comparisons are performed.  In doing so,
    expensive operations involved in computing L-norm distance
    functions between hyper-dimensional data can be avoided.

Biography:

 
  Dr Shen received Undergraduate Scholarship from Singapore Ministry
  of Eduction in 1996.  He obtained his BSc (with 1st class Honors)
  and PhD from School of Computing, National University of Singapore,
  in 2000 and 2003 respectively.  His research interests include
  autonomic computing, database, multimedia, P2P and internet
  applications.

  Heng Tao is supervised by Professor Beng Chin Ooi and rewarded as
  2001 Dean's Graduate Award Winner, School of Computing, National
  University of Singapore due to his meritorious performance. His
  journal and conference papers appeared mainly in database area
  (TKDE, ICDE 2004, ICDE 2003, etc) and multimedia retrieval (ACM
  Multimdedia, etc).

  Currently his research also inlcudes autonomic computing, data
  stream, bioinformatics, etc, and have several submissions on these
  areas.

Contact:

Professor Maria Orlowska, seminar host (maria@itee.uq.edu.au)
or Guido Governatori (ITEE seminar co-ordinator) (guido@itee.uq.edu.au)

ITEE seminar web page: http://www.itee.uq.edu.au/~seminar

[All seminars]