On-Line Signature Verification byDynamicTimeWarpingandGaussian Mixture Models

Handwriting signature isthemostdiffuse mean forpersonal identification. Lotsofworkshavebeencarried outtogetreasonable errors rates within automatic signature verification on-line. Mostofthealgorithms thathavebeen usedformatching workbyfeatures extraction. Thispaper dealswiththeanalysis ofdiscriminative powersofthe features that canbeextracted fromanon-line signature, how it's possible toincrease thosediscriminative powersby Dynamic TimeWarping asastepinthepreprocessing ofthe signal coming fromthetablet. Alsoitwill becovered the influence ofthis newstepintheperformance oftheGaussian Mixture Modelsalgorithm, whichhasbeenshownasa successfully algorithml foron-line automatic signature verification inrecent studies. Acomplete experimental evaluation ofthealgorithm baseon Dynamic TimeWarping andGaussian Mixture Models has beenconducted on2500genuine signatures samples and 2500skilled forgery samples from100users. Thosesamples areincluded atthepublic accessMCyT-Signature-Corpus Database.