3-Dimensional human head reconstruction using cubic spline surface on cpu-gpu platform

Abstract

This paper presents a CPU-GPU architecture for 3-dimensional human head reconstruction. The CT scan images of human head are processed to get the surface points, before spline surface are fitted to the points. In the process, a step is chosen to be executed on the GPU which is the cumulative sum calculation. This is because this step executes high processing time in the CPU. The developed architecture is tested on various numbers of surface points between 20 to 5000 points. Then, the performance of the developed architecture is measured based on execution time, speedup and granularity. The result shows that 3D human head reconstruction gives better performance by employing GPU in the process as compared to CPU, where the GPU execution time is 3.08 faster than the CPU. In terms of granularity, the computation time is 41.69 greater than the communication time. Therefore, the developed CPU-GPU architecture is suitable to be implemented for 3-dimensional human head reconstruction.