基于GPU并行的遥感影像边缘检测算法  

GPU-Based Parallel Edge Detection Algorithm for Remote Sensing Images

作　　者：陶伟东 黄昊[1,2] 苑振宇 杨柳[1,2] 王结臣 

机构地区：[1]江苏省地理信息技术重点实验室,江苏南京210093 [2]南京大学地理信息科学系,江苏南京210093

出　　处：《地理与地理信息科学》2013年第1期8-11,16,共5页Geography and Geo-Information Science

基　　金：国家科技支撑计划项目(2012BAH28B02)

摘　　要：针对数字图像处理领域中单个像元之间相互独立的特性,采用图形处理单元(GPU)的CUDA可编程模型,从粗粒度和细粒度双层并行结构出发,结合对Block和Thread的合理调度,对边缘检测Roberts梯度算子进行了算法的并行化设计和实现。实验结果表明,该算法充分利用了GPU强大的浮点并行计算能力,运行效率明显优于常规的边缘检测算法,对高分辨率遥感影像并行处理技术发展提供了有益借鉴。In light of the characteristic that pixels are independent from each other in the field of digital image processing, the pa- per adopted CUDA programmable model based on GPU, proceeded from coarse-grained and fine-grained parallelism, and com- bined reasonable dispatching of Block and Thread, attempting at parallel-designing and implementing Roberts operator, one of the traditional edge detection gradients. The experiment shows that the new algorithm is more efficient than ordinary edge de- tection algorithms. This algorithm makes full use of the powerful floating-point operation ability of GPU on parallelism and it serves as a good reference for developing parallel processing technology on high-resolution remote sensing images.

关 键 词：图形处理器 边缘检测 ROBERTS算子 并行化 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]