|
Content based image retrieval with relevance feedback:
A paradigm based on structure, color and texture
J. K. Aggarwal,
University of Texas, Austin:
Thursday, October 10, 2002, 4:30 PM,
Princeton University, Friend Center 004
Creation, use and retrieval of images permeate wide ranging disciplines today.
Surveillance, medical records, video summarization, satellite images and thermal images
are a few examples. Retrieval of images from a collection of images based on the content
of a query image is a challenging problem. At the University of Texas at Austin, we are
pursuing a number of projects on image and video processing. Traditionally, color and
texture have been the tools for retrieval of images from databases. We have added
structure derived via perceptual grouping to this list. This is accomplished without
segmenting objects of interest. The presentation will focus on deriving of structure via
perceptual grouping, and its use in the classification and retrieval of images. It will
further show that structure adds significantly to the efficiency of the retrieval of images,
and particularly for images containing man made objects. An added feature of our system
is the relevance feedback by the user. After the first response of the system to a query, the
user may evaluate the images retrieved by the system and initiate additional search. Our
system, available on the web, provides a hands on experience of retrieval and feedback.
The future use of our system in several applications including human motion,
surveillance and video summarization will also be discussed.
|
|
