Best (Machine Vision) Paper of the Decade

So if you were tasked with selecting the "most influential computer-science paper of the decade," where would you begin? (Let me make it easier -- you could start by eliminating anything that has my name attached to it.) Give up? So let's narrow the scope to, say, the most influential machine vision paper of the decade. Any ideas?

Okay, how about Gerard Medioni's 1996 paper "Using Computer Vision in Real Applications: Two Success Stories" presented at the International Association of Pattern Recognition Workshop on Application of Computer Vision, Tokyo, Japan in November 1996.

If this was your guess, then you're right on target. Medioni, who is chair of the Viterbi School department of computer science at the University of Southern California, has been awarded the "Most Influential Paper of the Decade" award from his peers in the machine-vision world for the classic paper he presented 11 years ago.

Back in the late 1990s, Gerard Medioni was engaged computer-vision research that led to, among other applications, techniques now used to integrate images into live video, so that viewers of a football game can see the line that must be crossed to get a first down.

The work grew out of algorithms Medioni created that let computers recognize objects in video, first looking for edges, then using those edges to define shapes, and then to recognize those shapes when viewed from different angles.

The breakthrough -- for which he is now being recognized -- came when he began to use the same algorithms to guide computers in creating images instead of interpreting them. In addition to the ability to insert images in live video, other applications included:

• A 3-D pen that lets visual designers create and mold 3-D shapes in a single step, rather than by the traditional means of using many 2-D renderings. The same device can "trace" existing objects in three dimensions to scan them into computers.
• A way to combine a minimal number of 2-D photographic views to create virtual 3-D objects, or even virtual environments.

Congratulations to Professor Medioni for a job well done.

Editor's Note: Professor Medioni has graciously made an early version of his paper available. You can read it here.

Posted by Jon Erickson at 10:41 AM  Permalink