Technical Papers

Video and Image Manipulation

Friday, 30 November 14:15 - 16:00 |  Peridot 201

Gaze Correction for Home Video Conferencing - Picture

Gaze Correction for Home Video Conferencing

We present a gaze correction system suitable for mainstream home video conferencing or podcasting. It uses inexpensive consumer hardware, achieves real-time performance and requires just a simple and short setup. Both the integrity and expressiveness of the face and the fidelity of the scene as a whole are preserved.

Claudia Kuster, ETH Zürich
Tiberiu Popa, ETH Zürich
Jean-Charles Bazin, ETH Zurich
Craig Gotsman, Technion - Israel Institute of Technology
Markus Gross, ETH Zürich

Discontinuity-Aware Video Object Cutout - Picture

Discontinuity-Aware Video Object Cutout

An efficient video cutout system that is capable of handling temporal discontinuities due to disocclusion, topology changes, etc. A novel directional segmenting classifier and a learning-based classifier combination method are proposed. High-quality results can be obtained with much fewer user interactions in comparison with existing systems.

Zhong Fan, Shandong University
Qin Xueying, Shandong University
Peng Qunsheng, Zhejiang University
Meng Xiangxu, Shandong University

Video Stabilization using Epipolar Geometry - Picture

Video Stabilization using Epipolar Geometry

We present a new video stabilization technique that uses projective reconstruction to treat jittered videos. This level of scene-understanding achieves the physical correctness of 3D methods yet avoids their lack of robustness and computational costs. We also describe a reprojection that accounts for non-stationary objects in a principled way.

Raanan Fattal, Hebrew University of Jerusalem
Amit Goldstein, Hebrew University of Jerusalem

Transfusive Image Manipulation - Picture

Transfusive Image Manipulation

We present a method for consistent automatic transfer of edits on one image to many other images of the same subject. We introduce novel, content-adaptive weight functions that enhance the Lucas-Kanade non-rigid alignment framework. Our method is robust to changes in view-point, illumination and subject arrangement.

Kaan Yücer, ETH Zurich
Alec Jacobson, ETH Zurich
Alexander Hornung, Disney Research, Zurich
Olga Sorkine, ETH Zurich