Integrating word level knowledge in text recognition

A text recognition system capable of correcting character-level confusions is developed. It incorporates word-level knowledge sources in the form of a dictionary and an envelope of the words. This speeds up the word recognition process by recognizing only crucial characters in the word. Also, the same knowledge sources are used to generate possible pairs of characters in a merged blob of characters, so that the blob can be segmented at proper positions. The system is capable of segmenting and classifying merged pairs of characters in the input text. A dictionary of the 9800 most frequently used words in English (taken from Brown Corpus) has been used