Technology Opportunities

Available Knowhow

Abstract

Given a video or a set of images taken from multiple viewpoints, surface mosaics attempt to include in a single 2D mosaic all the surfaces visible in the video, even when there is no single viewpoint from which all surfaces are visible. Surface mosaics are constructed by tracking feature points along the video, creating a track for each feature point. All tracks are embedded into the surface mosaic using low dimensional embedding, generating a mosaic where all feature points are visible.

Surface mosaics can use input frames from a camera having a general motion in 3D scenes, without the need to compute the scene structure or the camera motion. Feature tracks can partially overlap in time, or not overlap at all. Tracking errors and discontinuities are also addressed.

Since surface mosaics include all the surfaces visible in the input video, distortions are unavoidable. Distorted images that represent multiple views of objects are commonly generated by artists like Picasso and Hockney.

A surface mosaic can include several sides of an object in a single distorted 2D mosaic. This is in contrast to most mosaicing techniques which try to create undistorted mosaics, leaving out from the created mosaic many features visible in the input images.