Figure 4

Lumigraph geometry in compressive multi-view imaging. a Flatland-like illustration for collecting 1D images of an object in the 2D plane. At each camera position, all viewing directions may be considered. b Resulting 128 × 128 lumigraph for the ellipse-shaped object. Each row corresponds to a single "image". (In the real world, each image is 2D and the full lumigraph is 4D.) The lumigraph can be repeated for viewing from all four sides of the object. c Wedgelets provide a simple parametric model for local patches of a 2D lumigraph; only two parameters are needed to describe the orientation and offset of the linear discontinuity. d Wedgelet-based lumigraph reconstruction from M = 5 compressive samples of each image (row of lumigraph). e Scene geometry estimated using local edge positions/orientations in the reconstructed lumigraph. Each blue line connects an estimated point on the object to a camera from which that point is visible. The true ellipse is shown in red.