3-D object recognition from 2-d view

Abstract

This thesis presents two methods that can be used in recognizing 3-D objects from 2-D views. Those are the methods that use either the affine invariant Fourier descriptors and or the cross-ratio descriptors. The performance of affine invariant Fourier descriptors in recognizing 3-D objects is invariant under translation, rotation, scaling and shear distortion. However, it cannot cope with perspective distortion. On the other hand, the cross-ratio descriptors' performance is more than satisfactory under all those conditions including perspective distortion. This thesis also presents the simulation results applying the cross-ratio descriptors to the six simple 3-D objects which are individually different but in general quite similar.