Interactive Image Matting on Mobile Devices

The project involves researching statistically-based image matting techniques and interfaces for mobile devices. This would allow users to identify an object or other region in a photograph by indicating with a touchscreen parts of the image that are “inside” or “outside” the desired region. Statistical computer visions techniques will use the human input to find the best matting to isolate the desired object. As the user adds more information, the matting will automatically be updated. As this is intended for mobile devices, research will be required to ensure that the matting will work efficiently on low-power mobile processors. This work is related to the machine learning PhD work of the intern, and is valuable to the partner organization, as it is developing several photo-manipulation applications based on image matting, and improved techniques will ensure the applications are compelling on newer mobile devices.

Faculty Supervisor:

Dr. Nando de Freitas

Student:

Eric Brochu

Partner:

Pocket Pixels

Discipline:

Computer science

Sector:

Information and communications technologies

University:

University of British Columbia