A Novel Approach to Detect and localize the Text inNatural scene imag
Keywords:
Wavelet transform, Text Detection, Text Localization, Conditional random field (CRF)Abstract
The informationavailablein the naturalscene provides very importantcluesformanyimagebased application. So the
detectionandlocalization of textfromnaturalscene imageisimportant taskforcontent basedimageanalysis. In
thispaper,wedevelop anovelapproach todetectand localizetextsin naturalsceneimages. Thisproblem ischallengingduetothe
fact that text has different in size, alignment, orientation, style,complexbackground of imagesaswellasimage having the
lowcontrast. Initially,theRGBimageisconvertedintograyscaleimage andapplied to thelocal binarizationalgorithmfor the
segmentation. Afterthat,HaarWaveletTransform(DWT)isused which isfastestas compared withallwaveletsbecauseitsfilter
coefficientsareeither 1 or -1. It decomposeimageintofoursub band, oneisaverage andotherthreearedetailedwhichhelps to
find outtheapproximatelyconfidentregion.To filter out the non-text component, aconditional randomfield(CRF) is used
which appliedonconfidentregionimage. Finally,by usingsome predefinedcondition,thetextis obtainedin boundingbox.
