ICCV 2007

Eleventh IEEE International Conference on Computer Vision

Rio de Janeiro, Brazil, October 14-20, 2007

Professor Kanade's Speech

I feel most honored, and humbled, to receive the award that is named after Professor Azriel Rosenfeld, the godfather of computer vision. I first met Azriel 35 years ago in 1972 at the US-Japan Image Understanding Workshop that he organized in Kyoto. I still remember, vividly, that attendees included such luminaries in AI as Patrick Winston, Tom Binford, Raj Reddy, and Marty Tenenbaum. I was then a young, graduate student working on face recognition. I was given an opportunity to present my work in an unofficial “addendum” slot of the workshop. When I finished, Azriel gave me encouragement, and offered to edit my paper. Definitely that made my day.

I wrote my face recognition program in an assembler language, and ran it on a machine with 10 microsecond cycle time and 20 kB of main memory. It was with pride that I tested the program with 1000 face images, a rare case at the time when testing with 10 images was called a “large-scale” experiment. That was then. Since then, I have been fortunate enough to be given chances to work on many topics in computer vision and their applications. Motion, stereo, color, texture, line drawing, outdoor scene analysis, computational sensors, virtual and virtualized reality (as I like to call it), face and people detection, surveillance, robot vision for driverless cars, helicopters, and humanoids, medical robotics, and bio-cell images. I certainly had fun, too. With the broadcasting of EyeVision, a movie Matrix-like spinning football replay system, I could claim to be “The only professor who has ever appeared on the SuperBowl”.

The field of computer vision has come a long way: from curiosity to science; from toys to working systems. I feel that we are on the verge of an explosion of computer vision applications, as happened with computer graphics twenty years ago. We have to continue to develop technologies that work based on sound scientific and rigorous engineering disciplines. “Technologies that work” is the key phrase. Personally, I also like the recent revival of interest in the problems of scene analysis and recognition – THE original problem that the computer vision field set as an AI problem, and yet the problem that the field has been shied away for the last two decades because it is so difficult. But isn’t that the ultimate goal of computer vision? To understand and describe natural uncontrived scenes, indoor or outdoor. With greater computer power, newer tools, and younger talent, we can try again. In doing so, please be sure to learn from the past work – if you read papers from 70’s and early 80’s there are a lot of good ideas in them - of course, as well as bad ones.

From time to time people ask me “What is the secret of successful research?” I am not sure if I have had a lot of success, but maybe decent success. For that I would say, the best secret is to have good students and co-workers. I was extremely fortunate to have had many students who not only corrected my wrong ideas, but also turned them into great ideas. I am sure Professor Azriel Rosenfeld would have said the same thing for his many great students. I thank all of my past and present students and co-workers.

Finally I would like to tell all the students attending this conference, “Please help your professors”.