Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.752313
Title: Utilising path-vertex data to improve Monte Carlo global illumination
Author: Doidge, Ian Christopher
ISNI:       0000 0004 7425 4564
Awarding Body: Swansea University
Current Institution: Swansea University
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
Efficient techniques for photo-realistic rendering are in high demand across a wide array of industries. Notable applications include visual effects for film, entertainment and virtual reality. Less direct applications such as visualisation for architecture, lighting design and product development also rely on the synthesis of realistic and physically based illumination. Such applications assert ever increasing demands on light transport algorithms, requiring the computation of photo-realistic effects while handling complex geometry, light scattering models and illumination. Techniques based on Monte Carlo integration handle such scenarios elegantly and robustly, but despite seeing decades of focused research and wide commercial support, these methods and their derivatives still exhibit undesirable side effects that are yet to be resolved. In this thesis, Monte Carlo path tracing techniques are improved upon by utilizing path vertex data and intermediate radiance contributions readily available during rendering. This permits the development of novel progressive algorithms that render low noise global illumination while striving to maintain the desirable accuracy and convergence properties of unbiased methods. The thesis starts by presenting a discussion into optical phenomenon, physically based rendering and achieving photo realistic image synthesis. This is followed by in-depth discussion of the published theoretical and practical research in this field, with a focus on stochastic methods and modem rendering methodologies. This provides insight into the issues surrounding Monte Carlo integration both in the general and rendering specific contexts, along with an appreciation for the complexities of solving global light transport. Alternative methods that aim to address these issues are discussed, providing an insight into modem rendering paradigms and their characteristics. Thus, an understanding of the key aspects is obtained, that is necessary to build up and discuss the novel research and contributions to the field developed throughout this thesis. First, a path space filtering strategy is proposed that allows the path-based space of light transport to be classified into distinct subsets. This permits the novel combination of robust path tracing and recent progressive photon mapping algorithms to handle each subset based on the characteristics of the light transport in that space. This produces a hybrid progressive rendering technique that utilises the strengths of existing state of the art Monte Carlo and photon mapping methods to provide efficient and consistent rendering of complex scenes with vanishing bias. The second original contribution is a probabilistic image-based filtering and sample clustering framework that provides high quality previews of global illumination whilst remaining aware of high frequency detail and features in geometry, materials and the incident illumination. As will be seen, the challenges of edge-aware noise reduction are numerous and long standing, particularly when identifying high frequency features in noisy illumination signals. Discontinuities such as hard shadows and glossy reflections are commonly overlooked by progressive filtering techniques, however by dividing path space into multiple layers, once again based on utilising path vertex data, the overlapping illumination of varying intensities, colours and frequencies is more effectively handled. Thus noise is removed from each layer independent of features present in the remaining path space, effectively preserving such features.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.752313  DOI: Not available
Share: