Michal Irani
Video data provides a visual window
into the space-time world. It captures the continuous evolution of dynamic scenes
over extended regions in time and space. My research is focused on analyzing
the information contained in visual space-time volumes generated by video sequences.
In particular, I have been able to show that when moving away from image frames
and using all available information in entire space-time volumes, one can perform
tasks that are very difficult and often impossible to perform when only "slices"
of this information are used, such as discrete image frames, or discrete feature
points. This gives rise new powerful ways of analyzing and exploiting recorded
visual information from single and multiple video cameras.
My work is guided by real-world applications of video, with the aim of turning video data into a usable data-type. In particular, I have been focusing on applications such as action recognition, alignment and integration of information from multiple sensors in order to obtain enhanced visual sensing capabilities, alternative ways of visualizing recorded data, video synthesis and manipulation, rapid video search, video enhancement, video compression, and various surveillance applications.
Recent Publications
- [with D. Simakv, Y. Caspi and E. Shechtman] Summarizing Visual Data Using Bidirectional Similarity. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2008.
- [with O. Boiman and E. Shechtman] In Defense of Nearest-Neighbor Based Image Classification. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2008 .
- [with E. Shechtman] Matching Local Self-Similarities across Images and Videos. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2007.
- [with Y. Wexler and E. Shechtman] Space-Time Video Completion. IEEE Trans. on Pattern Analysis and Machine Intelligence (PAMI) 29 (3) 463-476, March 2007. (A shorter version appeared in CVPR'2004)
- [with O. Boiman] Detecting Irregularities in Images and Video. International Journal of Computer Vision (IJCV), special Marr Prize issue 74 (1) 17-31, August 2007. (A shorter version appeared in ICCV'2005)
- [with E. Shechtman] Space-time behavior based correlation -- OR -- How to tell if two underlying motion fields are similar without computing them? IEEE Trans. on Pattern Analysis and Machine Intelligence (PAMI) 29 (11) 2045-2056, November 2007. (A shorter version appeared in CVPR'2005)
- [with M. Blank, L. Gorelik, E. Shechtman and R. Basri] Actions as Space-time Shapes IEEE Trans. on Pattern Analysis and Machine Intelligence (PAMI) 29 (12) 2247-2253, December 2007. (A shorter version appeared in ICCV'2005)
- [with O. Boiman] Similarity by Composition. Neural Information Processing Systems (NIPS), Vancouver, December 2006.