提出了单位根假设检验可见光图像背景起伏平稳性的方法,理论上证明了时空域变化的广义平稳性。该方法采用ADF假设检验,从时域和空域分析背景灰度的变化,利用线性回归的方法进行协整性检验,从而得出结论。在平稳的基础上,使用相关长度刻划背景起伏的时空域相关性,得到两个结论。(1)一维指数模型的时间相关长度分析表明背景起伏具有强时相关性。并且这种强相关性集中体现在像素点灰度均值和相邻的有限帧(1~2帧)。因此提出低阶AR模型(1~2阶)对这种特性进行描述,得到的模型参数可用于图像背景估计。(2)空间相关长度分析证实了背景起伏具有强的空域相关性。并且图像可以分割成若干个不相关的区域。总的来说,背景起伏具有时空域的双重平稳性和强相关性。
Quantitativeanalysisvisibleimagebackgroundclutterstationarymethodviatheunitroottestwaspresented.Itisprovedintheorythattheclutterisgeneralizedtemporal-spatialstationary.ThemethodusedtheADFtesttoanalyzethecluttertemporal-spatialvariation,andusedline-regress-methodtotestcointegration,thenmadethatconclusion.Basedonthestationary,thecorrelativelengthwasappliedtodescribeitstemporal-spatialcorrelation,andthentwoconclusionsweremade.(1)ThetemporalcorrelativelengthisproposedbasedonIDexponentmodelanditshowsthattheclutterishighlytemporalcorrelative.Furthermorethecorrelationisincarnatedmostlybythesequencemeanandsomeneighborsfiniteframes(1~2frames).ThusloworderARmodel(1~2order)ispresentedtodescribethetemporalcorrelation.Andthesemodelcoefficientscanbeusedtoestimateimagebackground.(2)Thespatialcorrelativelengthanalysisprovestheclutterhighlyspatialcorrelative.Furthermore,imagescanbesegmentedtosomeareasuncorrelativewitheachother.Inaword,theclutteristemporal-spatialstationaryandcorrelativehighly,whichareimportanttobackgroundsuppressionanddetectingtargets.
Keyword(s)背景起伏,目标识别,目标探测,统计特性,图像处理;Backgroundclutter,Targetrecognition,Targetdetection,Statisticalcharacteristics,Imageprocessing