Experimental Performance Evaluation in Computer Vision

Professor Kevin Bowyer
University of South Florida

The need for sound experimental evaluation of computer
vision algorithms is more widely appreciated now than
it was in the past.  However, really are not yet any 
"standard" experimental performance metrics even for
basic and common computer vision tasks such as region 
segmentation and edge detection.  This talk will give
an overview of our experiences in developing peformance 
evaluation frameworks for cancer detection in mammography,
region segmentation in range images, and edge detection in
normal intensity images.  (It is intended that the talk
will be understandable to those not working in computer
vision.)