Introduction of statistical information in a syntactic analyzer for document image recognition

摘要: This paper presents an improvement to a document layout analysis system, offering possible solution Sayre's paradox ("a letter must be recognized before it can segmented; and segmented recognized"). improvement, based on stochastic parsing, allows integration of statistical information, obtained from recognizers, during syntactic analysis. We present how this fusion numeric symbolic information in feedback loop applied methods simplify description. To limit combinatorial explosion exploration solutions, we devised operator that optional activation the parsing mechanism. Our evaluation 1250 handwritten business letters shows method global recognition scores.