< Back to previous page


Advanced relighting algorithms for post-production

Book - Dissertation

In this dissertation, a set of algorithms are provided for relighting in a movie post-production environment. Lighting is a crucial factor in how an object appears to an audience. It can be used to hide certain parts of objects or to explicitly reveal them. Often, lighting controls the mood in scenes or conveys subtle emotions. It can occur that a particular scene configuration is not to the liking of a film director at some point in the production phase. Perhaps he would like to add some lighting or remove a disturbing glare that bounces of a chair next to a metallic object. In a regular movie production pipeline, the only way to proceed would be to perform a retake of that particular scene in which the configuration has been physically altered to reflect the director’s wishes. Depending on the scenario, this can be both costly and time-consuming. This dissertation will present algorithms that allow a director to avoid any physical recapturing of the scene in question. This has obvious cost advantages and could potentially permit unseen flexibility in movie production environments. The difficulty of designing relighting algorithms, lies in the appearance acquisition of the objects in a random scene. Not only are there several factors that contribute to the final shading of individual pixels, their data are also notoriously noisy, ambiguous or downright absent. Part of this work will be devoted to the explanation of the difficulties involved with estimating object geometry, material and illumination from nothing but images. As will be clear by the end of this dissertation, there are hard limitations on the ability to recover high-frequency details of these separate factors. On the other hand, not all use cases require the same amount of accuracy and some convincing results of relighting experiments will be demonstrated in this dissertation, corroborating the viability of our relighting algorithms for real-life applications.
Number of pages: 141
Publication year:2016
Keywords:relighting, post-production, wavelets, gpu, priors, optimisation