Hierarchichal attributed graph representation and recognition of handwritten Chinese characters

• Main Author: Ren, Ying, 1964-
• Other Authors: Memorial University of Newfoundland. Dept. of Computer Science
• Format: Thesis
• Language: English
• Published: 1991
• Subjects: Chinese characters > Data processing; Optical character recognition devices; Newfoundland studies; University of Newfoundland
• Online Access: http://collections.mun.ca/cdm/ref/collection/theses2/id/201015

Thesis (M.Sc.)--Memorial University of Newfoundland, 1991. Computer Science Bibliography: leaves 160-168. This thesis presents a system which is capable of recognizing handwritten Chinese characters. The hierarchical attributed graph representation (HAGR), a two-level graph, is introduced to describe the structural and statistical information of handwritten Chinese characters. The first level describes radicals and relations between radicals within a character, the second level describes strokes and relations between strokes in a radical. With HAGR, the recognition process becomes a simple task of graph matching. A cost function mapping a candidate to a model graph is introduced. This approach can tolerate the variations of HAGR which reflect the instablities and variabilities of handwritten Chinese characters resulting from different writing styles. Several rules have been used to re-arrange the order of the vertices of the graphs in order to avoid the combinatorial explosion inherent in graph matching. Based on HAGR, the model database is organized as a heterogeneous multi-way tree structure. For an input character, the search process can be divided into a number of simple and local decisions at different levels of the tree to find a corresponding model character in the database. The matching process is very efficient and accurate, and as well the system can acquire representations of characters by a learning process. Several HAGRs of samples of a character can be synthesized into a single HAGR of the character which can then be included in the model database. In addition, the learning process can update the models of characters with the HAGRs of their samples. The system is implemented in C on a MIPS/M-120 running RISC/OS (Version-3.1).