Interactive image segmentation for the diyorama mobile engine

Hololabs Studio Inc. is developing a mobile platform called Diyorama, which enables users to arrange images into interactive 3-D scenes with an aesthetic similar to paper craft and collage art. Users can include image from their photo libraries, social networks, or image search results in order to create customized playable scenes from existing content. The research project adds the ability to interactively extract foreground objects from unwanted background imagery in order to provide useful assets for building one’s scene. Computer vision techniques ranging from active contours to graph cut algorithms will be explored to see which is most effective for mobile platforms and user tests will be performed to ensure that the interaction is manageable in mobile environments, where small screens and multitouch input serve as the primary interface.

Faculty Supervisor:

Dr. Michael Langer

Student:

Fahim Mannan

Partner:

Hololabs Studio Inc.

Discipline:

Computer science

Sector:

Digital media

University:

McGill University