Task redistribution algorithm for contour extraction

A dynamic parallel algorithm for the active contour extraction with the varying number of control points has been developed. For the case where the execution time of tasks is not known in advance, the task scheduling algorithm speculates that a heavily-loaded processor might have more tasks in the near future and schedules less tasks to the processor than average. Thus, this algorithm keeps the balance of the parallel tasks over processors as fairly as possible and both the empirical and analytical evaluations show that the algorithm provides better performance than the balanced algorithm for the active contour extraction. The optimal scheduling parameter is also described for a special case. The analytical evaluation confirms the effectiveness of the algorithm and the experimental result shows that the algorithm provides up to 10% better performance than the balanced algorithm for the active contour extraction.