Prior, Context and Interactive Computer Vision

March 2, 2007
For many years, computer vision researchers have worked hard chasing the elusive goals such as "can the robot find a boy in the scene" or "can your vision system automatically segment the cat from the background". These tasks require a lot of prior knowledge and contextual information. How to incorporate prior knowledge and contextual information into vision systems, however, is very challenging. In this talk, we propose that many difficult vision tasks can only be solved with interactive vision systems, by combining powerful and real-time vision techniques with intuitive and clever user interfaces. I will show two interactive vision systems we developed recently, Lazy Snapping (Siggraph 2004) and Image Completion (Siggraph 2005), where Lazy Snapping cuts out an object with solid boundary using graph cut, while Image Completion recovers unknown region with belief propagation. A key element in designing such interactive systems is how we model the user's intention using conditional probability (context) and likelihood associated with user interactions. Given how ill-posed most image understanding problems are, I am convinced that interactive computer vision is the paradigm we should focus today's vision research on.

Category:

Education

Tags:

• CITRIS
• UCBerkeley
• Research
• Technology
• Engineering
• Science

[yezidi11]

i searched for algorithm i dont know how i got here, but good video i didnt watch all of it tho *****