Character Segmentation Using Side View Feature in Machine-Printed Optical Character Recognition
Min-Chul Jung · 한국정보기술학회논문지 · 2010
This paper defines four side views of a character and proposes a new character segmentation method of machine printed character recognition. The merged parts of touching characters generate different shapes of patterns from the primitive character patterns. However, the leftmost side and the rightmost side of touching characters will not be affected by the touching. The analysis of those side views gives the candidate single characters for touching characters, since a side view of each character is unique. Though characters in variable pitch have different widths, they can be grouped into thirteen classes. When it segments touching characters from the right to the left, the rightside view of touching characters provides a specific width among thirteen classes according to the candidate. In addition, both a segmental cost and a tangent cost are defined to find an optimal cut path. After touching characters are segmented with the width of a candidate character, the other side views of the segmented pattern are verified with those of a prototype character. The performance of the character segmentation has been obtained using a real envelope reader system, which can recognize address blocks in US mail pieces and sort the mail pieces. 100 mail pieces were tested. The experiment results have shown that the improvement was from 79% to 91% by the proposed character segmentation.