Optimized image sampling for view and light interpolation

Abstract

Accurate virtual reconstruction of real world objects has long been a desired goal of image-based computer graphics. Usually this involves a lengthy capture process where an object is photographed from different viewpoints and illumination conditions. Using this collection of input images, we can now re-render the object from any viewing angle or lighting condition. However, acquiring a dense sampling of both the lighting and view space is time consuming. We carry out an analysis on this combined lighting and view space to find the optimal sampling given a restricted image budget. We also analyze the order of interpolation and find that improved results are obtained by interpolating first in viewpoint and second in lighting, the reverse of the usual order.