A stroke-level wordnet for Farsi handwriting recognition

Language grammars and lexicons are essential tools for the post-processing word inference task in Handwriting Recognition Systems (HRS). In stroke-based HRS, input handwritten samples are recognized as members of standard written stroke categories. A stoke-level grammar is required to translate the classified strokes to words of a vocabulary. A wordnet is a tool to perform the word translation task at the least possible computational steps. In this paper, we develop a stroke-level wordnet for Farsi word recognition systems. The wordnet is obtained by parsing the vocabulary words through the Farsi stroke grammar rules. The wordnet, hence, includes lexical and grammar information, simultaneously; that reduces the cost of computation at the post-processing word recognition step. To handle the problem of infinitely many possible combinations of strokes in Farsi writing system and Persian language, we include multiple production rules per each stroke in the stroke grammar producing ambiguous explanations for out-of-dictionary words.