Optical recognition of chemical graphics

摘要

A prototype system for encoding chemical structure diagrams fromscanned printed documents is described. The system distinguishes astructure diagram from other printed material on a page image using sizeand spacing characteristics. It distinguishes line graphics from symbolsin an intermediate vectorization stage. Line information is mapped intoa connection diagram that represents atomic bonds. Atomic symbols areidentified by means of chemical drawing conventions and opticalcharacter recognition. The final coded output interfaces to conventionalchemistry software for database storage and retrieval, publishing, andmodeling