Abstract
In the production line for the automobile, the recognition of the frame number, stamped on the automobile chassis for production instruction or management, is indispensable. One problem of the alphanumerals stamped on the chassis is that the width and the depth of the stamped groove vary for each stamped character and position due to the variation of the stamping pressure of the stamping machine. Another problem is that a slow undulation is produced at an unspecified position of the stamped surface due to the strain near the stamped groove. When the stamped characters are imaged and binarized, adjacent characters connect and frequently one character separates into several pieces. Such connections and separations make it difficult to segment accurately the individual character regions. This paper considers the accurate segmentation of such stamped characters. The proposed method examines the extent of variation of gray level in each pixel‐line without binarizing the gray‐level image and classifies the characters by the differences of the variation. The result of experiment by the proposed method is as follows. For 11,000 stamped characters composed of 36 alphanumerals and two symbols, the rate of character extraction is 99.9 percent. For 1500 characters composed of 15 alpha‐numerals after character segmentation, the recognition rate is 99.5 percent without rejection. These recognition rates are higher than those in the conventional method, demonstrating the effectiveness of the proposed method.