Abstract:
In traditional computer graphics a scene to be rendered on a computer screen is described by a number of geometric primitives. However, construction of a geometric scene model can be a great deal of work and can be very expensive. Instead, image based modelling and rendering (IBR) employs pictorial information to describe and render a scene. Acquisition of images is straightforward and all properties of arbitrary objects are captured.
Theoretical foundation for IBR is the plenoptic function, which stores the entire illumination in a scene. A Light Field consists of a four-dimensional subset of the plenoptic function. In this way objects with arbitrary material properties and complex geometry are described. To obtain a Light Field, only a number of pictures of an object are sufficient. Unfortunately, a large
amount of data has to be stored in a Light Field to sample the plenoptic function with sufficient density.
The Waveletstream data structure presented in this work allows for progressive transmission, storage, and rendering of compressed Light Field data. For compression, the non-standard four-dimensional wavelet decomposition of the Light Field is calculated. The coefficients obtained are transmitted
progressively in a data stream and allow for display of the Light Field in different resolutions. To distinguish between fore- and background, the silhouette of the object is stored in the Waveletstream. In this way, not only integration of image based objects into conventional modelled scenes is facilitated, but also the compression ratio is increased. The Waveletstream allows for compression of Light Fields at a ratio of 1:100. Supplementary caching schemes allow for interactive decompression and display of Light Fields.
Furthermore, this thesis presents various examples for integration of image based data structures into interactive applications.