HANDWRITTEN CHARACTER-RECOGNITION USING A 2-LAYER RANDOM GRAPH MODEL BY RELAXATION MATCHING

A handwritten character recognizer is proposed in this paper. First, using components and strokes of a character as primitives, a 2-layer attributed graph is constructed to represent the character. Next, a relaxation matching technique is used in the learning stage to synthesize different attributed graphs, which are due to writing variations and belong to a same character, into a 2-layer random graph as a reference data model of the character. And the relaxation matching technique is also used in the recognition stage to match the attributed graph of an input character to the random graph of each reference character. Then, a similarity measure between an attributed graph and a random graph, based on whether the attributed graph is a possible outcome of the random graph, is presented in the recognition stage. According to the similarity measure, an input character can be classified into correct reference character. Experimental results are finally provided to show the effectiveness of the proposed approach. © 1990.