多图像同态滤波CPU与GPU并行算法分析与实现

《多图像同态滤波CPU与GPU并行算法分析与实现》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、万方数据STUDY栅¨_PLE—ENT^T10N0FCPU—GPUP^R^LLEL—ULT1．一I队GEH0—咖RPHICFILTER¨_G^LGORlT啊SLinageenhancementisimportantimagepreprocessingtechnologytostudy,howtoenhanceanimageefficientlyandquicklyisoneofthehotissuesinrecentyears．Homomorphicfilteringtechnologyisknownasakeyalgorithmofimag

2、eenhancementwhichisbasedOilthefrequencydomain．However,thealgorithmcontainsTheFourierTransformationandahugenumberofloops．Oncetheimageresolutionistoohigh，whenalargeamountofoperationsneedtobedone，theserialalgorithmwillbedifficulttoachievethereal-timeenhancement,SOtheparallelal

3、gorithmshouldbeusedtoimprovetheprocessingspeed．Therefore，thestudyingandthedesigningofTheHomomorphicfilteringparallelalgorithmformultiCoreGPU／CPUsystemshastheimportantpracticalsignificance．ThisarticledesignedandimplementedtheparallelHomomorphicfilteringalgorithmS-GPUTXLBform

4、ultipleimagesonthemulti-coreCPU／single-coreGPUsystembased011proceedingFouriertransformandinverseFouriertransformforparallelHomomorphicfilteringalgorithmsformultipleimagesbycallingGPUkernelfunctionlibrary．Basedontll她weusedtheindependenceofeachimageproceedingtask，thenweimplem

5、entedtheparallelHomomorphicfilteringalgorithmsD··GPUTXLBformultipleImagesonmulti·-coreGPUSystemsbyallocatingseveralimagestoseveralcoresofGPU．TheexperimentalresultsshowthatproceedthemultipleimagesHomomorphicfilteringparallelalgorithmS-GPUTXLBandD·-GPUTXLBbyusingthemulti-·cor

6、eCPU／GPUmachineCansignificantlyshortenthenmningtimeofprocessing．Thisarticleproceededwi廿1greedystrategy,andthenfindawaytotaskallocationforproceedingthehomomorphicfilteringparallelalgorithmformultipleimagesontheHybridheterogeneouscluster,throughcallingandproceedingoftheparall

7、elhomomorphicfilteringalgorithmsformultipleImagesonmulti-coreGPU／CPUsystemsateachnodeofthecluster．Atlast,wedesignedandImplementedtheParallelHomomorphicFilteringAlgorithmsD··GPUHCPAforMultipleImagesonMulti·-coreCPU／GPUSystemsontheHybridheterogeneouscluster．Theexperimentteste

8、dthealgorithmD—GPUHCPAwithdifferentnumberofthreadsII万方数据fromthethreadblockinGPU．We