| Using micro-nano optical structure is the newest3D holographic technology. Ithas broken the traditional bound so that we can get real3D images without any glassesor head sensors. The operation of extracting phase information is a key element ofit. Since the block processing and iterative operation in the whole procedure are quitetime-consuming, dealing with a group of9multi-view images in the size of1500×3300pixels with traditional serial algorithm costs more than10hours, so it’s hard toget into industrial production.CUDA can help us utilize the high parallelism physical structure to improve thecomputation performance greatly. It can play a role in the tasks of compute highlyparallel data. However, quantifcation and storing involve lots of logical judgmentswhich GPU is not good at because of the cache lack. So I’ve done a deep analysis ofthe parallel programming standards-OpenMP after the detailed studies of CUDA.In conclusion, this paper presents a model which combines CUDA and OpenMPtogether for acceleration. In this model, OpenMP in the high layer efectively supportsCUDA in the low layer. According to the proposed idea, we ve got the program hot,the most time-consuming code snippet, based on deep analysis of the process. Withthe proposed model, we can deal with the whole process through parallel method, andmake it more time-saving. We have compared our method with the existing serialalgorithm in many aspects in order to validate the feasibility of the proposed model.Comprehensive experiment shows that our method has achieved a good result:the efciency of parallel processing is200times higher than serial processing, and theaccuracy is also improved. Since the elapsed time of whole procedure dealing with agroup of9multi-view images in the size of1500×3300pixels has been extremelyreduced into5minutes, the efciency of phase information extraction has been greatlyincreased, which makes the technology closer to the requirement of industrial produc-tion. |
PDF Full Text Request