Accelerated volume rendering via spatial/spectral analysis

Volume rendering is an important and powerful technique for visualizing 3D datasets. However, the technique is computationally expensive, {dollar}Obigl(Nsp3bigr){dollar} for an N x N x N dataset. This thesis is concerned with the performance and image quality issues of volume rendering. Beginning with an analysis of the relationship between the computational expense of volume rendering and the image quality, the thesis shows how the two issues are related: high image quality is mainly the result of exhaustive and computationally expensive processing of the dataset. A hypothesis is developed that reducing the computational cost of volume rendering while retaining the image quality can be achieved by using an information metric to efficiently guide the rendering process.; This thesis presents a new algorithm for reducing the computational cost of volume rendering while retaining the image quality. The algorithm consists of a new analytically defined information metric and a hierarchical rendering algorithm for accelerating the rendering of volume datasets. The information metric is called Spectree, a compression code related to the Discrete Wavelet Transform. The computational cost of computing the Spectree is {dollar}Obigl(Nsp3bigr).{dollar} However, the computational cost of the hierarchical rendering algorithm is dependent on the information content of the dataset or the amount of information the user desires to be extracted from the dataset, O (information content). The user controls the algorithm by supplying a single parameter to specify the amount of information to be reconstructed from the volume. The choice of an information parameter is made easier or automated by a set of graphical tools which capture the trade-off between image quality and computational cost.