Light field rendering of under-sampled dynamic scenes by a plane sweep

Under-sampling in a light field causes ghosting artifacts to manifest in the synthesized images. The standard digital signal processing approach to removing these artifacts is by filtering the light field. However, this replaces the ghosting problem with image blur, which leads to an even greater loss of image detail. This dissertation introduces the plane-sweeping approach, which estimates scene geometry using stereo reconstruction techniques to compensate for the light field under-sampling. Furthermore, the geometry only needs to be estimated up to what is necessary for rendering an image free of ghosting artifacts. The plane-sweeping approach involves two ideas: reconstruction and rendering. The representation of the scene geometry allows both reconstruction and rendering to be achieved by image processing and warping. The visual hull and a library template shapes can also be easily adapted into this approach to improve the accuracy of reconstruction and quality of rendering.; This dissertation presents a hardware rendering framework for the plane-sweeping approach that can be realized by a real-time implementation on commodity graphics cards. Results will be demonstrated from experiments on both static and dynamic light fields that are under-sampled.