视频和图像处理中像素匹配运算的加速技术研究

《视频和图像处理中像素匹配运算的加速技术研究》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、国防科学技术大学博士学位论文视频和图像处理中像素匹配运算的加速技术研究姓名：谷会涛申请学位级别：博士专业：电子科学与技术指导教师：陈书明2011-04国防科学技术大学研究生院博士学位论文AbstractPixelmatchingmethodiswidelyusedincommunication,medicalcare,educationandmilitarydomain.Thetypicalpixelmatchingcomputationsincludemotionestimationofvideocoding,correlationma

2、tchingofobjecttrackingandrecognitionandsoon.Astherapiddevelopmentofvideoandimageprocessingapplications,theimageresolutionandframefrequencykeeptoimprove,andthethecomputationalcomplexityofpixelmatchinggreatlyincreases,presentingagreatchallengefortheperformanceofdigitalsigna

3、lprocessor.Therefore,theaccelerationresearchofpixelmatchingcomputationissignificantforimprovingtheprocessorperformanceandmeetingtheperformancerequirementofreal-timevideoandimageprocessingcomputation.Pixelmatchingisacomputation-intensive,data-intensiveandreal-timemethod.Ha

4、rdwareacceleratorisanefficientapproachtoimplementreal-timepixelmatchingcomputationMotionestimationandcorrelationmatching,whicharetheclassicpixelmatchinginvideoandimageprocessing,arestudiedinthispaper.Andseveralaccelerationtechniquesonpixelmatchingcomputationareexploredtoi

5、mprovetheflexibilityandperformanceofmotionestimationandcorrelationmatchinghardwareaccelerator,optimizethefastalgorithmimplementedbyhardware,andincreasetransmissionefficiencyandflexibilityoftheinterfacebetweenhardwareacceleratorandprocessor.Thedetailedperformanceanalysisan

6、devaluationforeverytechniquearecarriedoutonourDSPplatform.Themaincontributionsandinnovationsofthisthesisareasfollows:1)Amultiplesearchcentersmotionestimationalgorithmsuitableforhardwareimplementisproposedtospeedupthecomputation.Theproposedalgorithmisbasedonmulti-searchcen

7、terspredictionanddynamicsearchrangeadjustment.Themulti-searchcenterspredictionanalyzesmotionvectorsofspatialandtemporaladjacentblocksandpredictsmultiplemotionvectorsforthecurrentblock.Comparedwiththetraditionalmotionvectorprediction,theproposedpredictionmethodcanimproveup

8、to12.9%predictionaccuracy.Accordingtothecountandmagnitudeofthepredictivesearchcenters,thesearchr