Giga-scale multiresolution volume rendering on distributed display clusters
HCIV'09 Proceedings of the Second IFIP WG 13.7 conference on Human-computer interaction and visualization
| Hi-index | 0.00 |
Direct volume rendering is a powerful visualization method that has been used for many years. Rather than displaying geometrical primitives and polygons, each discrete unit of space in the volume, known as a voxel, is rendered in the image. This results in the ability to see the internal structures and materials in the volume, and the gradual transitions between different substances within an object. This is especially useful for biomedical data, where organs have very complex structures and tissues rarely have clear surfaces and boundaries, instead they are composed of materials mixing together, and objects occluding one another. Multi-dimensional transfer functions, with the sampled data values and their derivative magnitudes as parameters, are used here to assign opacity levels to voxel points. A technique is presented in which a user can interactively select control points to create a smooth 3D spline curve function as the transfer function. However, for one inexperienced with a particular dataset, it can be difficult to manipulate higher-order curves with good precision. Therefore, a data clustering analysis on the histogram of intensity, and first and second order derivative magnitudes is used to automatically assign control points for a piecewise spline curve of the multi-dimensional transfer function. This method provides more intuitive control over visualizing the dataset to the user's specific needs. As new advancements in scanning technology continue to progress, higher and higher resolution images are produced which result in much larger scale volume data sets to be rendered. A restructuring of the dataset into an octree structure with wavelet decompositions of the leaves is presented, which provides a volumetric spatial subdivision and a multi-resolution hierarchy of the data. This technique allows for loading large datasets quickly into memory, and for making the most efficient use out of the available graphics card's limited texture memory. The results of these methods are shown as images produced with interactive frame rates, using the described direct manipulation interface to find multi-dimensional transfer functions. The efficiency of the octree and wavelet format is shown by comparing the data loading times with that of the standard cross-sectional image format.