Multiple viewpoint rendering

This paper presents a method for efficiently rendering a sequence of regularly spaced perspectives of a scene. The algorithm presented, called Multiple Viewpoint Rendering (MVR), exploits perspective coherence by rendering epipolar plane images. The computer graphics camera geometry is constrained to produce epipolar plane images with linear features. Transformation and shading operations can be performed once per image sequence instead of once per view. Geometric position, color, and texture coordinates can be computed using linear interpolation. Both view independent and dependent shading algorithms are supported. Both one and two-dimensional grids of perspectives can be rendered. Details of a hardware-accelerated implementation of a multiple viewpoint polygon renderer are given. 1 Introduction Most of the time, photography captures information about the world as seen from a single viewpoint. A still cameras single lens records a scene as it appears at one instant, while a movie cam...