Blurestimationisrequiredinimageprocessingtechniquessuchasauto-focussing, qualityassessmentforcompressedimagesandimagefusion.Inthispaperamulti-scale localblurestimationmethodisproposed.Weusetheenergyofapairofquadrature filterswithfirstandsecondderivativesofaGaussianatseveralscalesasitsconstituents. A newstrategyforanalyzingtheextremaofenergyacrossscaleisproposed.Comparing to themethodswhichusejustaGaussianfirstderivativekernel,asmallernumberof scalesneedstobeprocessed.Alsoourmethodyieldsonlyoneresponseatthecentroidof line-shapefeaturesatapositionindependentofthescale.Thisisincontrasttoother methodswhichyieldmultipleresponsesatscaledependentpositions.Weevaluatedthe methodforsyntheticandrealimagesfromtheLIVEdatabase.Dependingondetailsof theimage,theproposedmethodisseveraltotensoftimesfasterincomparisonwith usingjustaGaussianfirstderivative.Theaccuracyoftheblurestimationisfoundtobe thebestorsecondbestwhilecomparingwith16differentmethodsforGaussianblur.In addition,theperformanceisstillgoodformotionblurredimages.
