Thursday, November 21, 2013 - 4:00pm to 5:00pm
KEC 1005

Speaker Information

Feng Liu
Assistant Professor
Department of Computer Science
Portland State University


The increasing availability of cameras means that one is always on hand to capture an interesting moment. However, a good video is difficult to capture. One of the most obvious differences between professional and amateur level video is the quality of camera motion. This talk will describe our work on re-cinematography that allows a user to transform their
hand-held shaky videos to have the appearance of an idealized camera motion, such as a tracking shot.

Re-cinematography addresses two challenges in improving camera motion: where and how to move video content to have a good camera motion. The first problem is difficult because no existing motion model both captures camera motion well and is easy to compute. 3D model is difficult to compute while 2D model cannot handle video with parallax. Re-cinematography estimates eigen feature trajectories that can model the apparent motion of a video well and can be computed online. The second problem is challenging because it requires rendering a novel view. Existing methods for novel view synthesis cannot handle scene dynamics and parallax simultaneously. Re-cinematography solves this problem using a content-preserving warping method that creates each output frame from only its corresponding input frame according to the target motion. These two techniques together lead to a re-cinematography system that offers the first method that both achieves high-quality camera motion and is practical enough for consumer applications. This work has led to commercial products, such as the Warp Stabilizer in Adobe After Effects CS 5.5 and Premiere Pro CS 6.0, and has been widely used in film, such as Martin Scorsese's Oscar-award-winning movie Hugo and David Fincher's The Girl with the Dragon Tattoo.

This talk will also give a brief overview of the computer graphics and vision research at Dr. Liu's lab, such as computational cinematography, computational stereography, 3D video visualization and interaction, and computational aesthetics.

Speaker Bio

Dr. Feng Liu received his Ph.D. in Computer Science from the University of Wisconsin-Madison in August 2010 and joined Portland State University as an Assistant Professor in September 2010. His research interests include computer graphics, computer vision, and multimedia. He received the Best Thesis Award in the Computer Sciences Department at UW-Madison. He was named as one of the IBM Watson Emerging Leaders in Multimedia '08. He received the Best Paper Runner-up Award at ACM Multimedia 07 and the Best Paper Honorable Mention Award at ACM CHI 13. He was awarded the 2012 Intel Early-Career Faculty Honor Program Award.