The first generation of transitive stereo-metric satellites in China, TH-1 Satellite, is able to gain stereo images of
three-line-array with resolution of 5 meters, multispectral images of 10 meters, and panchromatic high resolution images
of 2 meters. The procedure between level 0 and level 1A of high resolution images is so called relative radiometric
correction (RRC for short). The processing algorithm of high resolution images, with large volumes of data, is
complicated and time consuming. In order to bring up the processing speed, people in industry commonly apply parallel
processing techniques based on CPU or GPU. This article firstly introduces the whole process and each step of the
algorithm – that is in application – of RRC for high resolution images in level 0; secondly, the theory and characteristics
of MPI (Message Passing Interface) and OpenMP (Open Multi-Processing) parallel programming techniques is briefly
described, as well as the superiority for parallel technique in image processing field; thirdly, aiming at each step of the
algorithm in application and based on MPI+OpenMP hybrid paradigm, the parallelizability and the strategies of
parallelism for three processing steps: Radiometric Correction, Splicing Pieces of TDICCD (Time Delay Integration
Charge-Coupled Device) and Gray Level Adjustment among pieces of TDICCD are deeply discussed, and furthermore,
deducts the theoretical acceleration rates of each step and the one of whole procedure, according to the processing styles
and independence of calculation; for the step Splicing Pieces of TDICCD, two different strategies of parallelism are
proposed, which are to be chosen with consideration of hardware capabilities; finally, series of experiments are carried
out to verify the parallel algorithms by applying 2-meter panchromatic high resolution images of TH-1 Satellite, and the
experimental results are analyzed. Strictly on the basis of former parallel algorithms, the programs in the experiments are
written by parallelizing the serial version. Numerical results show that, the parallel algorithms proposed in this article
can effectively bring up the processing speed of RRC for high resolution images of TH-1 Satellite, and, with its stable
running and excellent image processing, meet the need of application. For three kinds of images, such as images with
large amount of cloud, with little differences of ground features, and with good climate and plentiful ground features, the
average acceleration rates could reach 2.484, 2.539 and 4.159 respectively.
|