Overview of Appearance Based Methods in Computer Vision

Abstract

Appearance based methods turned traditional computer vision approaches to object recognition upside down. Instead of going from image to 3D model and matching that model to models of know objects, appearance based methods stay in the 2D image domain. Objects are modelled based on how they can appear in images, and this is learned directly from images; it is not necessary to have any explicit knowledge of the shape or re ectance properties of the object. This technical report presents an overview of appearance based methods in computer vision.