New strategy allows computerized search of digital image databases
By Douglas Page
University researchers have come up with a new way to automatically
sort, classify, and retrieve digital images. Their method promises
faster, more accurate database searches for applications such as
radiologic data mining.
The new approach, called SIMPLIcity, looks at
images in a manner similar to the way people look at images. Just as a
person shown a picture of a horse can extract features characteristic
of horses and then identify other pictures that contain horses, this
new computer-based system approaches images, said its developer, James
Z. Wang, Ph.D., an assistant professor of information sciences and
technology at Pennsylvania State University who specializes in medical
informatics.
Using wavelets and a novel technology called
integrated region matching (IRM) that performs region-based image
similarity comparison, the system retrieves relevant images from any
image database or from the Web on the basis of automatically derived
image features or content.
Images are represented by a set of regions,
roughly corresponding to objects, characterized by features reflecting
color, texture, shape, and location properties, according to Wang. IRM
evaluates overall similarities between images, incorporating properties
of all the regions in the images using the region-matching scheme.
"The system processes each image in the database,
extracts key features, indexes the features in the feature space, and
retrieves images based on the feature comparison scheme -- all of which
enables users to search for visually related images from massive image
databases," Wang said.
SIMPLIcity has been validated on a database of
about 200,000 general-purpose images and an archive of more than 70,000
digital pathology images.
"The system is targeted to retrieve electronic
medical images, although it also works well for general-purpose
photographs," he said. "Medical image retrieval is a lot more demanding
in terms of accuracy."
Other potential applications include education,
biomedicine, crime prevention, the military, commerce, entertainment,
and Web image classification, Wang said.
Image retrieval techniques in commercial use rely
mostly on key words or descriptions. While these text-based approaches
are accurate and efficient for limited databases, it becomes
prohibitively expensive to manually input descriptions on larger scales
such as image databases containing an astronomical number of
observations, or radiographs. The new approach eliminates the need to
input textual information.
To view a demonstration of SIMPLIcity, go to http://wang.ist.psu.edu/IMAGE.
POST YOUR COMMENTS
|