Abstract: If you want to snap a picture of a bald eagle in the zoo, a Little League baseball game, or some nice city landscape from a tall building, chances are you will have something obstructing your view. In this talk I will describe work we have done in this space over the past couple of years, developing computational techniques for taking clean pictures through obstructions such as window reflections, fences and glare. I will briefly describe our obstruction-free photography work (SIGGRAPH'15), where we used camera motion and a generative model to separate and remove the obstruction layer from the main scene, then show how we have productionized it at Google to help users take glare-free pictures of prints with their phones (PhotoScan; ranked among the top 10 apps of 2016 by Fast Co and Mashable). I will also cover recent work we have done on designed, intentional obstructions -- visible image watermarks (CVPR'17) -- showing how these watermarks can be removed automatically (flagging a security flaw), and how they can be made more effective.Joint work with Ce Liu, Mike Krainin, Tianfan Xue, Tali Dekel, and Bill Freeman.

Bio: Michael Rubinstein is a Senior Research Scientist at Google, Cambridge MA. His research is at the intersection of computer vision and graphics, and focuses on areas in computational photography and image/video processing. He received his PhD in Computer Science from MIT in 2013, advised by Prof. Bill Freeman. Prior to joining Google he spent a year as a Postdoctoral Researcher at Microsoft Research New England.