In this talk, I will discuss the work I have done over the past few years in scene and object modeling using both geometric and image-based techniques. This includes model completion based on both the incomplete geometry and calibrated images when available. As a techniqe for image and geometry processing, I will discuss the implementation of spherical harmonic transforms and convolutions on the GPU, with the aim of generating an overcomplete spherical wavelet representation of an image or geometric data.
I will then discuss how a biorthogonal spherical wavelet can be used to extract parameterized face models from noisy stereo data. Finally, I will talk about scene modeling and novel view synthesis using spherical disparity layers.