Keyword spotting in degraded document using mixed OCR and word shape coding

This paper presents a new way for keyword spotting in degraded imaged document. Two prevalent word indexing, OCR and word shape coding, are combined compactly based on the recognition confidence evaluation. The basic procedures are as follows. First, OCR candidates are used for OCR indexing. Second, a new stoke feature and convex-concave feature of word are adopted for word shape coding. Furthermore, an intelligent indexing based on recognition confidence is introduced, which is adaptive to image quality. Finally, an inexact matching is used for word spotting. A collection from NLM, including 1553 scanned imaged documents, is used to evaluate our method. The results confirm the validity of our method.