Graphic symbol recognition using a signature technique

The authors present a noise-robust and general-purpose method for the recognition of graphic symbols in line-drawing images. The method assumes that a noise-free symbol database is available, and that both the model symbols and the processed image have been broken down into elementary structures. Recognition is based on the hypothesis-and-test paradigm. The detection of an elementary structure allows us to use a signature filter, to hypothesize and then recursively reduce a set of matching symbols. Hypothesized symbols are then verified by exploiting distance transform and distance measurement