An accelerated shape based segmentation approach adopting the pattern search optimizer

Abstract

All known solutions of the shape based segmentation problem are slower than real-time application requirements. In this paper, the problem is formulated as a global optimization problem for an energy objective function with several constraints. This formulation allows the use of the global optimization solvers as a solution. However, this solution will be slow as it requires the evaluation of the objective function for several thousand times. The objective function computation is one of the critical factors that affect the time needed to reach a solution. The authors implemented two accelerated parallel versions of the solution that integrates the objective function and the pattern search solver. The first uses a GPU accelerated implementation of the objective function and the second uses a CPU parallel version which is executed on several processors/cores. The results of the proposed solution show that the GPU version has substantial speed compared to other approaches.