并行图像复原与超分辨处理系统的设计与实现

Implementation of parallel image restoration and super-resolution processing system

为实现海量大尺度图像的复原与超分辨算法的实时处理,设计了图像复原和超分辨并行处理系统,对该系统的技术方案、工作原理、关键算法、硬件结构和并行技术进行了研究。根据技术方案,介绍了系统的工作原理及解模糊、去噪和超分辨等关键算法。对数字信号处理(DSP)和机群两种体系结构进行了分析和比较,结果表明机群更适合于大规模并行处理,给出其设备选型的原则。提出了基于PPF结构的并行算法模型和基于MPI+OpenMP混合结构的多层次并行与优化技术。最后,对算法在DSP和计算机系统上的处理效果和处理速度进行了实验和分析,给出了系统规模和性能的预测,确定了关键参数即处理器数目的选择依据。实验结果表明,该系统可将处理时间由2 700 s降低到29.39 s,处理后的图像的清晰度、对比度和分辨率显著提高,满足应用需求并具有一定的通用性。

A parallel processing system was developed to realize the real-time restoration and super-resolution processing of a huge amounts of large size images after investigating its technical scheme, working principle, key algorithms, hardware architectures and parallel technology. The working principle of the system and key algorithms of deblur, denoise, super-resolution were presented based on the technical scheme, and two kinds of the architectures,Digital Signal Processing（DSP） and Cluster systems, were analyzed and compared. It is concluded that the Cluster system is more large scale parallel processing. A parallel algorithm model based on the PPF architecture and a multi-layer parallel optimization technology based on the hybrid architecture of MPI＋OpenMP were introduced. The processing effect of the proposed algorithm and the processing speed on the DSP and computer were analyzed. Following by prediction of the scale and performance of system, How to choice the number of processors was also discussed. Experimental results show that the parallel system can reduce run time from 2 700 s to 29. 39 s, and can provide improved images with better sharpness, contrast and resolution.