Contents

Various Image Based Rendering (IBR) techniques have been proposed to reconstruct scenes from its images. Voxel-based IBR algorithms reconstruct Lambertian scenes well, but fail for specular objects due to limitations of their consistency checks. We show that the conventional consistency techniques fail due to the large variation in reflected color of the surface for different viewing positions. We present a new consistency approach that can predict this variation in color and reconstruct specular objects present in the scene. We also present an evaluation of our technique by comparing it with three other consistency methods. Acknowledgements I would like to thank Mark Stevens, my thesis advisor, whose support, insight and direction made this work a success and for providing me with the all the resources and funding to work on this project; Micheal Gennert, my thesis reader, whose guid-ance and encouragement helped me to formulate my ideas and for proof-reading thi