Optimizing an inverse warper

Traditionally, in the field of computer graphics, three-dimensional objects and scenes have been represented by geometric models. Image-based rendering is a powerful new approach to computer graphics in which two-dimensional images, rather than threedimensional models, act as the underlying scene representation. Warping is an image-based rendering method in which the points in a reference image are mapped to their corresponding locations in a desired image. Inverse warping is a displaydriven approach to warping. An inverse warping algorithm first constructs a continuous representation of a reference image. Then, it maps reference-image points to desiredimage points by sampling the continuous reference image with rays sent out from the desired image. Inverse warping has major advantages over forward warping methods in image recon struction and warping from multiple reference images, but implementations of the algo rithm have proved to be very slow. This paper presents two optimizations for inverse warping -a hierarchical implementation and an improved clipping method -that improve inverse warping's performance dramatically. Thesis Supervisor: Leonard McMillan Title: Assistant Professor