图像恢复的高效并行算法及关键技术

首次从并行处理的途径分析了能产生高恢复质量，但具有高计算复杂性的图像恢复算法BNM的并行性，并对影响该算法并行效率的关键问题，提出了有效的解决方案：①采用条状重叠的数据分配方案，减少了并行处理中的通信量；②给出了不同读取策略的内部实现模型，分析了不同读取策略对I／O带宽产生的影响，提出了能够获得高I／O性能的读取策略；⑧提出了降低通信量的“关键位通信”方法．综合运用上述策略，设计并实现了高效的并行BNM算法．理论分析和实验表明，该并行BNM算法具有很高的加速比、并行效率及很好的可扩展性，是解决图像恢复实用性的有效途径。

AN EFFICIENT PARALLEL ALGORITHM FOR IMAGE RESTORATION AND ITS KEY TECHNIQUES

Best neighborhood matching (BNM) is an error concealment algorithm to achieve high quality image restoration. However, BNM needs intensive computation, which restricts its real application. In this paper, a parallel BNM is proposed. Several critical techniques have been developed to obtain high performance. These techniques include overlap stripe data distribution, reading strategy, and communication strategy. Theoretical analysis and experimental results show that the parallel BNM has good speed up and scalability so that it can provide an efficient way for image restoration.