Seite - 196 - in Document Image Processing

J. Imaging 2018,4, 32 typesofXMLfileshavebeengenerated,basedonthe informationcontainedintheglobalXMLfile, one for thedetectiondataset and the other for the recognitiondataset. ThedetectionXMLfile is providedat the line level foreachframe. Figure9adepictsapartof thedetectionXMLfileofFrance24 TVchannel. Oneboundingbox isdescribedby theelementRectanglewhichcontains the rectangle attributes: (x,y) coordinates,widthandheight. Therecognitionground-truthfilesareprovidedat the line level foreachtext image. TheXMLfile iscomposedof twomarkupsections:ArabicTranscription andLatinTranscription. Figure9bdepictsanexampleofaground-truthXMLfileandits textline image. Figure8.ApartofaglobalXMLannotatingavideosequenceofAljazeeraTV.Thisfigurecontains ground-truth informationabout three text-boxes fromatotalof17. a b Figure9.ExampleofAcTiV2.0specificXMLfiles: (a) apartof thedetectionXMLfileofFrance24TV; and(b) a recognitionground-truthfileanditscorrespondingtextline image. 196