The  Weizmann  Institute  of  Science
Faculty of Mathematics and Computer Science

Vision and Robotics Seminar

Iddo Drori
School of Computer Science
Tel Aviv University

will speak on

Example-Based Style Synthesis

Recently a number of example-based image synthesis techniques have been developed (e.g., image analogies, texture synthesis, super-resolution, separating style and content), but none of them deals with the synthesis of new style without $\underline{\rm natural \ correspondence}$.  The meaning of style and content is quite intuitive and a human artist can easily reproduce a novel image, repainting it with a desired style. The artist is well trained and has a deep knowledge of what distinguishes the style from content. For a machine to do this is a fairly complex task. In this talk I will introduce an example-based image synthesis technique that interpolates and synthesizes new styles for a given input image. A globally non-linear model of the training set validates that an image can be soundly reconstructed, and its style and content parameters estimated. By altering the style parameters, novel images are synthesized that contain the same content with an interpolated style. I will show some preliminary results of on-going research.

Joint work with Daniel Cohen-Or and Hezy Yeshurun.

The lecture will take place in the
Lecture Hall, Room 1, Ziskind Building
on Thursday, April 25, 2002
at noon