Abstract
Volume rendering is a well-known technique for visualizing scalar fields in medical imaging and scientific data. But extensive computation time required is one of its major draw backs. The shear-warp volume rendering algorithm proposed by Philippe Lacroute and Marc Levoy can render and image almost in real time, provide that the raw volume data have been preprocessed into run-length code. However, to prepare a run-length code for a given volume data requires much more time than rendering an image. To cope with this problem, we proposed an efficient parallel algorithm for preparing the run-length code in a distributed computing environment. The basic idea of our parallel algorithm is to distribute both control and data among multiple computers to achieve parallelism. We use the master-slave model for effective distributed control and interleaving subdivision method to achieve data parallel. Furthermore, the techniques such as task queue and task stealing are also addressed and incorporated into our algorithm for load balance and better performance. To demonstrate the feasibility and the efficiency of the algorithm, we implement and test it on the three different distributed computing environments with Parallel Virtual Machine (PVM) as our application interface (API) to the distributed computing environment. The experimental results show that our parallel algorithm can achieve satisfactory speed up factors. Finally, same parallel computation techniques are exploited to construct a multiple viewing system, which can automatically display a sequence of volume images according to a predefined view direction within a reasonable time.