ASIFTMatchingAlgorithmBasedOnAdaptiveContrastThresholdYouZHAIDepartmentofPostgraduateManagementTheAcademyofEquipmentCommand&TechnologyBeijing,ChinaHYPERLINK"mailto:*youyou1952@sina.com"youyou1952@sina.comLuanZENGTheKeyLabAcademyofEquipmentCommand&TechnologyBeijing,Chinazengluan@sina.comAbstract—SIFTkeypointsarerejectedifthecontrastislessthangiventhreshold.Thethresholdhastobeselectedfordifferentimagesmannulayandthereforecannotachieveautomaticimagematching.InordertoimprovetheaccuracyandautomationofSIFTbasedimageregistrationalgorithms,aSIFTmatchingalgorithmbasedonadaptivecontrastthresholdisproposed.Firstly,computeSIFTcontrastthresholdthroughnormalizedimageentropyandthenthethresholdisusedtorejectlowcontrastSIFTkeypoints.Secondly,theNearestNeighboralgorithmisusedtoobtaininitialmatchingsetaccordingtocommonmatchingthreshold.Thirdly,theobviousmismatchesarerejectedusingthehistogramofmainorientationdifferenceoftheinitialmatchingsetandthencaculatethecontrolparameterkofmathcingthreshold.Afterthat,matcheswith8smallestratiosareselectedtosolvetheinitialperspectivetransformparametersfromthetransformedimagetorefferenceimageandthencalculatethemaximumerrorσfromthe8transformedkeypointsofthetransformedimagetothe8correspondingkeypointsofthereferenceimage.Finally,computetheerrorfromthetransformedkeypointsofthetransformedimagetothecorrespondingkeypointsofthereferenceimageforallinitialmatches.Matchesarerejectediftheerrorsarelargerthan3kσ,thentherefinedmatchingsetisobtained.Theexperimentalresultsshowthattheproposedmethodcaneffectivelyextractcorrectmatchesandismorerobusttolowcontrastimagematching.Moreover,thecontrastthresholdandthematchingthresholdarecomputedautomaticallyfordifferentimage.Keywords-imageprocessing;SIFT;automaticmatching;entropy引言SIFT算法[1][2]不仅对图像缩放、平移和旋转具有不变性，而且对光照变化以及复杂的仿射和投影变换也具有部分不变性，比较适合用于存在透视变换的图像配准[3]。在实际系统中，经常由于光照的不均匀造成图像灰度的过度集中，从而造成图像对比度的降低。笔者发现SIFT算法提取的特征点的数目随着图像灰度分布的集中而减少，当图像灰度分布集中到一定程度时，SIFT算法鲁棒性会大大降低。因此SIFT算法虽然具有一定的光照变化不变性，但是仍然不能够很好满足低对比度图