Template-Free Word Spotting in Low-Quality Manuscripts
Huaigu Cao, Venu Govindaraju · Advances in pattern recognition · 2006
As the OCR technique is not yet adequate for handwritten scripts with large lexicon, word spotting has been introduced as an alternative to OCR. This paper proposes a novel approach to word spotting that, instead of matching features of the word image to features extracted from predefined templates, uses the estimated posterior probability as the output of well trained classifier for spotting. Gabor features are extracted from gray scale image in order to yield higher performance on degraded, low quality document image. Keywords: Keyword spotting; OCR.