Exploring Defocus Matting: Non-Parametric Acceleration, Super-Resolution, and Off-Center Matting

 
Neel Joshi Wojciech Matusik Shai Avidan Hanspeter Pfister William T. Freeman
University of California, San Diego MERL MERL MERL MIT
IEEE Computer Graphics and Applications, Special Issue on Computational Photography



(a)


(b)


(c)


(d)

(a) A prototype of the A-Cam, marked with a red rectangle, attached to a standard consumer level camera, marked with a green rectangle. The A-Cam is a collection of three low quality cameras that share a single center of projection, using a tree of beam splitters. (b) Original high-def image from a consumer camera. (c) The super resolution alpha matte. (d) Super resolution alpha premultiplied foreground image.

 

Abstract

 

Image matting for scenes with arbitrary, unknown backgrounds is an important operation in photography and film production. A promising approach is defocus matting, which pulls mattes from video captured with co-axial cameras with different depths of field and plane of focus. The method is attractive as it is fully automatic and passive.We explore defocus matting and introduce several new techniques that push the envelope of what can be achieved with the method. We introduce a non-parametric sampling method to accelerate the video matting process from minutes to seconds per frame. We present a super-resolution technique, which can efficiently bridge the gap between high-resolution video cameras and mattes from low-resolution cameras. Finally, we show how to pull mattes for an external high-resolution camera that does not share the same center of projection as the low-resolution cameras used to capture the defocus matting data.

 

Paper

Video