Week 03 Colloquium Summary
David Banks


    Experimental Performance Evaluation in Computer Vision
                           Kevin Bowyer
                   University of South Florida


Dr. Bowyer advocated standardizing the way that image-analysis algorithms 
are evaluated. He offered the example of mammogram interpretation. Different 
algorithms are presented with a set of images; each algorithm decides 
whether an image contains a cancerous region, and it decides the severity 
of the cancer.

The decisions are compared to those made by human experts. The results are 
displayed graphically using an ROC curve, plotting true positives against 
false positives.

He has compared a few algorithms to each other, and has made the data 
available on the Web. His goal is to improve the performance of computer
vision algorithms by making them compete under fair, objective conditions.