Transcript imaging_presentation

The Collaborative Imaging Grid
A collaborative research environment enabling Researchers
to learn from images when computer vision cannot
Paul Javid, Kurtis Heimerl
What is the problem with computer
vision?

Computer Vision is getting better at seeing patterns in images
 Face Recognition for Security (Is this person Kurtis?)
 Traffic Monitoring (Are there many cars on the road?)

But we are no where near the capabilities of the human eye!
 Medical research (Does this lung have lung cancer?)
 Spyware detection (Is this image spyware?)
 Requires Human Pattern Recognition!
The Solution

Create a research tool allowing researchers to aggregate
human knowledge about digital images.
 Images are coupled with metadata
 Metadata contains additional information about the images
 Researchers have access to “database” which contains all
images and their coupled metadata
 Creates the “world-wide-web” of human knowledge about
images.
Example application scenario

100 Cancer Researchers have knowledge about what a
cancerous lung looks like.
 In order for such information to be shared ubiquitously:
 When researcher identifies cancerous lung image, they add
that image and its metadata to database
 Metadata contains information about patient history, age,
etc.
Researcher then may analyze other images and contribute
to their metadata
 Allows other researchers to search, group, edit, and
categorize images in database based on their metadata,
thereby furthering cancer research
How are we going to do this?
 Clients have access to database via a web-browser.
 Outputs DHTML
 Server side uses GNU’s Tomcat webserver and the Spring
Framework.
 Toolkits implemented in Java!
 Metadata stored as XML files.
 Velocity toolkit to generate HTML for client
 Clients log in via web browser to Tomcat webserver which
contains database of images and coupled metadata (XML Files).
Server exports HTML for clients viewing
Conclusion
 Obvious demand for Collaborative Imaging Grid in various
disciples - biotechnology, computer vision, etc.
 Easy to Implement
 Well known implementation language (Java)
 Easy integration between XML metadata, webserver, and
client side browser.
 We have done this before!