Trevor Darrell a distinguished professor from UC Berkeley gave a very interesting invited talk yesterday at IUI’09 on the role of image processing in making user interfaces more intelligent.
He elaborated on the state-of-the-art of image recognition. There are essentially two categories of this field: 1) instance level recognition and 2) category level retrieval. The first category is almost a solved problem, while the second one sill causes headaches to scientists: how could computers recognize that a bar stool and a rocking chair belong to the same category?
Trevor also showed a few new cutting edge technologies that are starting to change our everyday life. For example, the SnapTell application for smartphones can recognize snapshots of book covers, CDs, etc. made with your cameraphone and find the product on the internet. The goal of this program is of course commercial, to sell the items that you are looking for. But such applications can also enable you to find out more about e.g. a new movie, based on a snapshot of its poster or even a TV advertisement.
This idea seems relevant to the work of my colleague, Michael Farrar, at Project54. His photo tagging application could be further improved to enable location, object or even suspect recognition (in the long run) for police officers.
Oszkar Palinko
No comments:
Post a Comment