As I have shown in this chapter, the previous fifteen years have seen great strides in volume rendering. Although commodity graphics hardware still does not directly support volumetric primitives, several techniques utilize graphics hardware to perform some or all of the volume rendering. However, as the capabilities of graphics hardware continue to improve, we must continue to look for new ways to take advantage of them.
In addition, many approaches exist for computing the volume rendering integral. However, they all have flaws. Averaging luminance is fast but can be quite inaccurate. We can perform linear interpolation of luminance accurately, but it is far slower than other ray integration methods. Pre-integration computes the volume rendering integral outside of the rendering loop, so the system is capable of performing accurate computations without affecting the rendering rate. However, the size of the lookup table limits the accuracy. Furthermore, we must rebuild the table every time the transfer function changes, which can be often. None of these algorithms can perform computations that are both fast and accurate for changes in viewing position and transfer function.