Touching characters are responsible for the majority of errors in the machine-printed character recognition. This paper defines two character segmentation costs, which are a cutting cost and a tangent cost. The cutting cost is defined as how many blac...
Touching characters are responsible for the majority of errors in the machine-printed character recognition. This paper defines two character segmentation costs, which are a cutting cost and a tangent cost. The cutting cost is defined as how many black pixels should be apart in order to segment a pattern. The tangent cost is defined as how many black and white pixels a segment line passes by consecutively. A font classifier provides a candidate character of touching characters. The width of the candidate character gives a reference line to segment touching characters. A few neighboring columns of the reference line gives a possible segment area. The cutting cost and the tangent costs can find an optimal segmenting path to segment touching characters and improve the performance of character segmentation.