banner research
By J. Kniss, G. Kindlmann, C. Hansen
manHeadSinusY3

This page documents the evolution of our volume rendering project, named Simian. As volume rendering goes, this system is quite a departure from the way this process is typically approached. Most direct volume renderings produced today employ one-dimensional transfer functions, which assign color and opacity to the volume based solely on the single scalar quantity that comprises the dataset. Multi-dimensional transfer functions, however, are an effective way to extract specific material boundaries and convey subtle surface properties. However, finding good transfer functions is hard enough in one dimension, let alone two or three.

memri2
A typical MRI slice

Simply put, volume rendering is a graphics technique that makes images from 3D data. The data typically comes from CT or MRI scans of the human body, but can also come from computer simulations of physical phenomena like fluid flow, explosions, molecular interaction, and weather patterns. A volume is made up of voxels. Voxels are to volumes what pixels are to images. We think of a voxel as having eight corers, like a cube, which contain the data. We can estimate what new data values inside the voxel should be using interpolation. A transfer function maps data values to colors. We can generate an image by sampling voxels at regularly spaced intervals, then mapping these values through the transfer function, and finally projecting them to the screen.

The key to insightful direct volume rendering is the effective use of transfer functions that map the data values to opacity and color. Transfer functions are fundamental to volume rendering because their role is essentially to make the data visible. By assigning optical properties like color and opacity to the voxel data, the volume can be rendered with traditional computer graphics methods. To date, transfer functions have generally had only one-dimensional domains, meaning that the 1D space of scalar data value has been the only variable to which opacity and color are assigned, though often, there are features of interest in volume data that are hard to extract and visualize with 1D transfer functions. For instance, many medical datasets created from CT or MRI scans contain a complex combination of boundaries between multiple materials. The overlapping ranges of data values spanned by these boundaries mean that a transfer function based on data value alone will be unable to isolate the individual boundaries. Higher dimensional transfer functions can permit the visualization of subtle variations in properties of a single boundary, such as its thickness or sharpness.

img40
This image shows how the data value, f(x), its first derivative, f'(x), and its second derivative, f''(x), change as we pass through the boundary between two materials. Notice how f(x) moves smoothly from a low data value to a high one. f'(x) goes from low to high and back to low. f''(x) goes from 0 to a high value, then through zero to a low value, and finally to 0 again. We know that a data value is at a boundary when f'(x) is high and f''(x) is zero.
In our paper "Interactive Volume Rendering Using Multi-Dimensional Transfer Functions and Direct Manipulation Widgets"(to appear in the IEEE Visualization 2001 Conference), we describe a novel 3D interface that we developed to deal with finding boundaries and setting transfer functions to show them. The problem with traditional volume rendering is that finding boundaries may be difficult if you only have a transfer function to play with. As you can see from the illustration above, the relationship between data values and boundaries is complicated. Our interface lets you probe at the volume and have it tell you what the relationship is. In addition, our widget allows the user to automatically set the transfer function based on the feature being probed. This makes the process of transfer function specification considerably more intuitive.

probe1
A probe for poking around in the volume
clip1
A clipping plane can be used to cut away parts of the volume. It also shows you the data mapped through a different transfer function so that you can see what's going on. You can also click on it and have the transfer function widget show the value that you selected.
tf
This is our transfer function widget. It only shows you two dimensions of the 3D transfer function at a time. The vertical dimension is first derivative and the horizontal is the data value. The triangles are widgets which let us set good transfer functions. The balls strung together in the middle of the widget are how we show the data values pointed to by the probe or the clipping plane.

While our research aims to explore the importance and power of multi-dimensional transfer functions, our main contributions thus far have been two techniques that make volume rendering with multi-dimensional transfer functions easier and faster. To resolve the complexities inherent in a user interface for multi-dimensional transfer functions, we have first introduced a set of direct manipulation widgets that make finding and experimenting with transfer functions an intuitive, efficient, and informative process. In order to make this process genuinely interactive, we must next explore the fast rendering capabilities of modern graphics hardware, especially three-dimensional texture memory and pixel texturing operations. Together, the widgets and the hardware form the basis for new interaction modes that guide the user towards transfer function settings appropriate for their visualization and data exploration interests.

Examples

Spheres

We use spheres as test volumes because they have easily identifiable features, predictable surfaces, and they are very simple to generate.

veryfirst3Dtf
This was the very first volume rendering using a 3D transfer function. 3-02-2001
firstLight
This was the first volume rendering using a 3D transfer function with lighting. 3-07-2001

Sinuses

Sinuses are hard to visualize because they either show up as bone or skin. We can show them in isolation because we have better discrimination of data values with 3D transfer functions.

headLevoyTF2
The sinuses of a female (red)
manHeadSinusY3
The sinuses of the Visible Male

Shading

Shading is a natural way to convey shape and curvature information.

manHeadSkullNoShade
Visible Male's skull without shading
manHeadSkullShade
Visible Male's skull with shading

Thickness

The first derivative helps us discriminate between surfaces which are made of the same material, but have different thicknesses.

blade-plain
The turbine blade is made out of one type of metal.
blade-fancy
Its surfaces, however, have different thicknesses. Red is thin, blue is thick.

Painting

We can use the probe and clipping plane to paint into the transfer function.

manHeadWidget
Point the probe at an interesting surface, like the skin.
manHeadPaint
Now paint the data value into the transfer function.
manHeadProbeIN
Look at the soft tissue just by pointing.
headMarrow1
Examine the bone marrow by painting. Use a widget to show the skull.

More Spheres (eye balls)

manHeadEyeBall4
The eyeballs of the Visible Male
manHeadEyeBallTF
The Transfer Function that captured them. The eyeballs are in the very small triangle.

Parallel

Some times a volume is too BIG to render on a single graphics card. TRex, is a parallel hardware volume renderer which renders smaller parts of a large volume on several graphics cards, and then pieces them back together.

CGASheep3CGASheep4 CGASheep2CGASheep1
This is what the smaller chunks look like.
CGASheep0
This is what they look like when they are pieced back together.

This is an MRI of a sheep heart. Notice the valve between the atrium and ventrical (this is the left ventrical). The clipping plane helps expose it. Notice the fine filaments which attach it to the heart wall (papillary muscles). This is an example of 1D transfer functions. They can show materials well, but boundaries are rather difficult to extract.