Akcelerace real-time light field renderingu na GPU

This thesis describes original proposals regarding light field rendering, focusing, compression, streaming, and methods for its optimal usage on 3D displays.  Light field rendering is a fast image-based way to render otherwise computationally demanding synthetic or even real-life scenes. An input set of images that captures a scene is used to synthesize a novel view without complex 3D reconstructions. The main scientific contribution of this thesis is a proof of a hypothesis. The hypothesis states that state-of-the-art light field rendering methods can be outperformed by a novel method in terms of visual quality, memory requirements, and time performance. The novel method proposed in this thesis enables the usage of light fields in real time without visual quality degradation and excessive hardware requirements. The hypothesis is proven experimentally and supported by reference implementations and published papers. The thesis also contains other novel proposals that serve as supporting materials for the main method. These proposals address the most crucial issues in light field rendering. The main contribution and other proposals are intended to open a way to the usage of light field assets in the gaming and film industry. This thesis provides an overview of existing state-of-the-art methods, the hypothesis with experimental proof, and a description of proposed applications of light field principles on 3D displays.

light field, 3D display, GPU, image-based rendering, multiple views, plenoptic function