Amnon Shashua

Abstract

This paper investigates the differences — conceptually and algorithmically — between affine and projective frameworks for the tasks of visual recognition and reconstruction from perspective views. The study is made by first proposing an affine framework for perspective views, captured by a single remarkably simple equation, which is based on a viewer-centered invariant we call relative affine structure. Via corollaries of the main result we make connections to previous work and show that Euclidean, affine and projective structure representations can be obtained by simple specializations and generalizations of our main results.