“4137X˙CH07˙Akerkar” — 2007/9/8 — 11:21 — page 284 — #22
284 CHAPTER 7 Web Content Mining
queries that take advantage of spatial relations, such as “Find images with object 5 above
object 9.”
For retrieval-by-content with images, it is important to keep in mind that we can practically
work only with a restricted notion of semantic content, based on relatively simple “low-level”
measurements, such as color, texture, and simple geometric properties of objects. There are
numerous regular distortions in visual data: translations, rotations, scale variability, nonlinear
distortions, and so on. The human visual system is competent to manage certain distortions
without difficulty.
7.6.1 Demonstration of a Multimedia Information Retrieval System
IBM’s Query By Image Content (QBIC) search technology helps you locate artwork using
visual tools at the website of the State Hermitage ( www.hermitagemuseum.org/html
En/
index.html), which occupies six magnificent buildings situated along the embankment of the
River Neva, right in the heart of St. Petersburg, Russia. By using the URL www.hermitage
museum.org/fcgi-bin/db2www/browse.mac/category?selLang=English you can find artwork
by selecting colors from a palette or by sketching shapes on a canvas. Requesting all artwork
with comparable visual attributes can further refine the existing search results.
The image information retrieval system allows you to conduct tasks such as finding a
Gauguin masterpiece simply by recalling the organization of his subjects or locating a Da
Vinci painting by searching for its predominant colors. You can visually search for artwork
using tools that an artist would use. Due to copyright restrictions placed by Hermitage, we
cannot include the images from the website here. For an overview of the QBIC searches,
take a look at the animated demonstrations at www.hermitagemuseum.org/fcgi-bin/db2www/
qbicSearch.mac/qbic?selLang=English.
The QBIC Color Search allows us to specify colors and locates two-dimensional artwork in
the Digital Collection that match. The colors are selected from a palette that contains a spec-
trum of colors. The search can then be executed after defining the proportions. Visit the QBIC
Color Search Demo to view a step-by-step demonstration of this search at www.hermitage
museum.org/fcgi-bin/db2www/qbicColor.mac/qbic?selLang=English.
In addition, you can use the QBIC Layout Search for specifying geometric shapes. Using
geometric shapes, you can arrange areas of colors on a virtual canvas to approximate the visual
layout of the work of art that you are looking for. QBIC interprets the virtual canvas as a grid
of colored areas. This grid is then matched to other images stored in the database. A QBIC
Layout Search Demo that illustrates a step-by-step demonstration of this search can be found
at www.hermitagemuseum.org/fcgi-bin/db2www/qbicLayout.mac/qbic?selLang=English.
EXERCISES
1. (Project) Perform the same keyword searches using three different search engines. De-
scribe the number of documents retrieved. Combine the top ten documents into a docu-
ment collection, go through these documents and mark them as relevant or nonrelevant.
Calculate precision and recall for the search results. Compare the differences of the top
ten pages found by each. Envisage why such differences exist.

Get Building an Intelligent Web: Theory and Practice now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.