一种适用于机群系统的细粒度遥感图像镶嵌并行算法

Fine-grained parallel algorithm for remote sensing image mosaics for cluster system

为得到全局色调一致的无缝镶嵌图像 ,提出了一种细粒度遥感图像镶嵌并行算法。该算法通过维护一个“双缓冲队列”和采用“任务动态选择”算法进行任务分配 ,极大地减少了通讯开销和任务等待 ,而且图像的色调均衡化和重采样都可融入到镶嵌算法中并行执行。还提出了一种基于不规则边界的改进λ插值图像融合算法 ,用以进行图像重叠部分的无缝拼接。对镶嵌算法性能的实际测试和分析结果表明 ,该算法在机群系统上获得了近似线性的加速比和良好的视觉效果。

The combining of image mosaics is an important part of remote sensing image processing. A fast fine grained parallel algorithm was developed to produce seamless image mosaics with global grayscale coherency. The algorithm uses a double buffer queue and a dynamic task selection strategy to greatly reduce the communication overhead and task waiting time. The equalization and resampling procedure can be easily embedded into the parallel model. The algorithm also includes an improved image merging algorithm which supports irregular boundary image merging. Experiments to test the efficiency and effectiveness of the algorithm show that the new algorithm has approximate linear speedup and provides very good visual results.