Use this URL to cite or link to this record in EThOS:
Title: Customised light probes and inverse lighting methods for relighting
Author: Calian, Dan
ISNI:       0000 0004 7965 0590
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
This thesis tackles the inverse lighting problem of computer graphics. Specifically, the thesis introduces novel methods of estimating the distant lighting environment which illuminates a real-world scene given, as input, a single photograph of a known object found or placed within the scene. Two immediate applications of the inverse lighting techniques introduced herein are in: 1) augmented reality where virtually rendered content is integrated with a live view of the real-world scene in real-time and, 2) in film post-production where virtually rendered content is composited with previously shot footage in an offline process. Achieving realistic integrations of virtual and real content requires, at a minimum, accurate geometric and plausible radiometric calibration (in the form of omnidirectional lighting estimation). This thesis contributes strictly to radiometric calibration but does make use of existing geometric calibration methods in experimental validations. We first introduce a custom voxel "shading probe" which enables approximate global illumination rendering of arbitrary voxel models by lightmapping simple linear combinations of the probe's captured faces. We then extend the "shading probe" idea to cater for arbitrary, non-voxel, 3D models. We propose a spherical-partitions probe, whose shading, at a few designated surface points, encodes optically-computed basis-convolved incident illumination. These shading values can then be directly used as basis coefficients for lighting in a precomputed radiance transfer framework. Following the same line of investigation, we posit the use of the human face as an outdoor lightprobe. To achieve this goal, we model incident lighting by the latent space of a convolutional autoencoder trained on skies as well as by a statistically-constrained pre-specified parametric sky model. Finally, we introduce a novel theory of fast convolutional sparse coding on the sphere with inverse lighting applications. We express our theory in a computationally efficient form by leveraging state-of-the-art spherical harmonics theory for decomposing anisotropic spherical signals as weighted combinations of rotated zonal harmonics.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available