Seite - 161 - in Document Image Processing

J. Imaging 2018,4, 39 Table3.Classificationresults forElliptical featuresetwithMLPClassifier. Class Class A B C D E F G H I J K L R A 355 29 49 48 0 25 3 4 42 10 6 29 2 B 2 550 0 7 0 1 32 0 0 0 0 8 27 C 27 0 479 8 1 19 2 11 31 1 8 13 32 D 32 9 0 514 0 13 23 0 3 2 4 0 66 E 66 1 2 1 441 42 4 7 10 20 4 2 96 F 96 3 6 15 6 397 16 4 19 12 14 12 55 G 55 13 7 54 6 17 402 1 3 19 22 1 25 H 25 0 2 3 0 26 0 491 28 10 4 11 7 I 7 0 23 3 33 8 7 3 493 10 4 9 0 J 0 0 1 0 16 5 2 2 9 553 6 6 2 K 2 0 16 7 1 7 12 0 2 7 546 0 8 L 8 22 1 0 6 6 20 13 6 9 1 508 0 Now, theconfidencevaluesprovidedto theclasses forevery inputdatabytheclassifiersonthe threesetsof features formthe input for theclassifiercombinationprocedures. Theconfusionmatrix resultingfromtheMajorityvotingprocedure ispresented inTable4.Anoverallaccuracyof95.6%is achievedonthisdatasetcontaining7200samplesdividedequallyamongthe12scriptclasses. It isseen thatDevanagariscripthasgot the leastaccuracyandgetsconfusedwithTeluguwhereashighaccuracies areshownforManipuriandOdiaandBangla. Borda count algorithmgives anaccuracyof 93.5%which is an increase of 2.1%over thebest performingindividualclassifier. Itprovides thehighest recognitionrate forDevanagariamongall the combinationschemesandgoodaccuracies forotherpopularscripts likeBanglaandOdiaandhence canbe thepreferredchoice forwideusage. The trainableversionof thealgorithmwithweightsbased onoverallaccuracyof theclassifiers improves theresults further. The increase is2.9%withsatisfactory results forscripts likeTelugu,KannadaandUrdu. Theaccuracyfor theGurumukhiscript remains low irrespectiveof theweights. Theresultsarepresented inTables5and6. Thesimplerulesat themeasurement level tocombine thedecisionsprovidegoodresults in the presentwork. Thesumruleattainsanaccuracyof 97.76%withalmost close toperfect recognition forUrdu,Gurumukhi andRoman. The product rule andmax rule have accuracies of 95.73% and 94.60%respectively.Highestaccuracy is foundforOdiascriptwhereasproduct rulesuffers incaseof Gurumukhiandmaxrule incaseofDevanagari. Theresults for theelementaryrulesofcombinationare tabulated inTables7–9. Sumruleoutperformsallother rulebasedcombinationapproaches in thisworkandtestifies the resultspresentedbyKittler et al. mentioned in [38]bybeing lessprone tonoise anduncleandata. TheDStheoryresults combine theresults, twoata timeandthenall three together. Theclass-wise performancebasedBPA,whichoutperformstheglobalperformancebasedBPA,hasbeenimplemented for themulti-classifier combinationusing theDS theory [45]. The rule applied for this process is quasi-associativeandhence theresultsofcombiningtwosourcescannotbecombinedwith the third. Therulehas tobeextendedto includeall the threesources together. Results for thecombinationof the classifier resultsonHOGandElliptical features,MLGandElliptical features, and,HOGandMLG featuresarepresented inTables10, 11and12respectively. Thecombinationresult includingall the threesourcesof information isgiven inTable13. There isno improvementshownbythecombinationof theresults fromMLGandHOGfeature sets. But when the Elliptical feature set is involved in the combination process there is much improvementover theparticipatingclassifiers.Overallaccuraciesof91.2%and97.04%areachieved bycombiningsourceshaving78.1%and79.4%accuraciesand91.4%and79.4%accuracies respectively. So, improvementsof6%and10%are foundbyapplying theDStheoryofevidence. Combiningall three,anaccuracyof95.64%,morethan4%over thebetterperformingclassifier isseen. Inboththe schemesall the script classeshaveaccuraciesover90%andwithalmost 100%accuracy for certain 161