Language Independent Word Spotting in Scanned Documents

Large quantities of scanned handwritten and printed documents are rapidly being made available for use by information storage and retrieval systems, such as for use by libraries. We present the design and performance of a language independent system for spotting handwritten/printed words in scanned document images. The technique is evaluated with three scripts: Devanagari (Sanskrit/Hindi), Arabic (Arabic/Urdu) and Latin (English). Three main components of the system are a word segmenter, a shape based matcher for words, and a search interface. The user gives a query which can be (i) A word image (to spot similar words from a collection of documents written in that script) or (ii) text (to look for the equivalent word images in the script). The candidate words that are searched in the documents are retrieved and ranked, where the ranking criterion is a similarity score between the query and the candidate words based on global word shape features. For handwritten English, a precision of 60% was obtained at a recall of 50%. An alternate approach comprising of prototype selection and word matching, that yields a better performance for handwritten documents is also discussed. For printed Sanskrit documents, a precision as high as 90% was obtained at a recall of 50%.