A new approach for recognition of unconstrained handwritten numerals

In this paper, a new approach for recognizing unconstrained handwritten numeral is proposed. First, smoothing and coloring processes are used to smooth the shape and to obtain the number of components of the observed numeral. Then the crossing method, starting from the top left corner, searches for the crossing number for each line of the character horizontally and a stroke changing sequence is derived from the crossing number sequence. Finally, a set of primitives derived from the left and right profiles of the external character contours are defined. After each primitive of the external contours of a character is obtained, a pattern extraction procedure is proposed to extract the basic patterns from the primitive changing sequence. By using the primitives and the basic patterns, each numeral is reconstructed and represented by a 7/spl times/7 binary matrix. Based on these features, a decision tree is established for recognition. Experimental results show that a high recognition rate is obtained.<>