Two-Stage Word Spotting Scheme for Historical Handwritten Devanagari Documents

Massive group of valuable historical documents are existing in many national libraries. There is a need for creation of digital libraries that helps in information retrieval. Word spotting is one of the methods that spots all occurrences of a given query word image from the documents, which is a substitute method of a complete transcription of documents. There are many challenges in spotting words in historical handwritten Devanagari documents like variant writing style, deep degradation and cursiveness present. In this paper, a novel two-stage word spotting method in historical handwritten Devanagari documents is presented. In the first method, gradient, structural and cavity (GSC) features are used for representing handwritten words and in second method, a bag of low-level discrete features are used for representing the handwritten word images and by combined hybrid approach, word spotting is performed in handwritten documents. Extensive experimentation is performed on handwritten Devanagari Shankaracharya’s manuscripts and the results for proposed approach are promising in terms of accuracy and efficiency.