Déjà vu

An algorithm for zooming in on repeating elements in images and videos can help reconstruct missing information

But what about visual information? A family photo, for instance, or a video of last summer’s vacation? Prof. Michal Irani of the Faculty of Mathematics and Computer Sciences has discovered that, here too, if one zooms in on small enough parts of the picture, one can discern many tiny visual elements that repeat themselves again and again inside the picture. Similarly, in video clips, tiny repeating space-time elements can be identified.

 

Irani creates algorithms that enable computerized systems to identify such tiny, repeating visual elements and use them to predict visual information that is missing from a picture or a video clip. She can reconstruct visual information about the scene that the camera has not been able to record (e.g., due to resolution limits or obstructions). The identification of repeating visual elements further enables compact “visual summaries” of pictures and video clips; it helps distinguish objects from background, and suggests new ways of compressing visual data.

 

Prof . Michal Irani's research is supported by the Citi Foundation.

 

 

---

 

 

Mixed Media | Guy Goldstein
Pencil and graphite on paper, 2011

 

Can one represent reality with a grid? Can one arrange and organize many “objects” in a large chest of drawers? Guy Goldstein doubts, for the moment, that this is the way things really work. Thus, possibly, the reality relayed through his grid-like images rebels against the rigid chart with lines that loop back on themselves. This subtle interference eventually breaks up the straight lines and even tries to break through the boundaries of the drawing.