Near Real-Time Parallel Image Processing using Cluster Computers

The utility of cluster computers (Beowulfs) for efficient parallel image processing of large-scale mosaics and stereo image correlations is presented. The parallelization of existing serial software using the message passing interface (MPI) is described. Almost perfect reduction of the required wall-clock time with increasing number of CPUs is reported. This true parallelization approach differs from the use of clusters as serial task farms in that it reduces the total time required by individual tasks and enables near real-time interactive data analysis. The reduction of required processing time by over one order of magnitude furthermore enables the addition of processing steps to monitor the quality of stereo image correlation tiepoints. The algorithm for such quality assessment and the impact with respect to the choice of correlation algorithms is discussed.