提出了一种人民币光学字符识别 (OCR)过程中号码区域快速定位的新方法。算法首先利用固定阈值法对纸币图像进行二值化, 然后从二值纸币图像的左边缘和下边缘选取一定数目的样本点, 根据非磨损残缺区域两边缘上向量内积为零的特性, 筛选出位于边缘非残缺区域内的样本点, 从而排除边缘磨损残缺的干扰。再根据这些样本点用最小二乘法拟合直线, 以确定纸币的倾斜角和左下角点坐标。最后对图像做旋转变换, 同时根据旋转公式得到纸币图像矫正后左下角点的坐标, 并以角点为起点偏移固定距离提取出号码区域, 这使得提取号码区域时间开销在微秒级, 图像矫正与号码定位几近同时完成。实验结果表明, 该方法具有较好的鲁棒性和较高的时间效率, 在纸币号码识别系统中具有很好的应用前景。

A novel approach to region location of serial numbers in Optical Character Recognition (OCR) of RMB is proposed. Firstly, a binary image was obtained by applying fixed threshold segmentation to the banknote image. Secondly, some of sample points of left and lower edges were selected by using sequential scanning. Thirdly, due to the characteristic that the dot-product of vectors from two edge lines on non-damaged areas is zero, the sample points of edges on non-damaged areas were sifted by filtering sample points on damaged areas. So, a slope angle and a lower-left corner can be got by using least-square line regression on these points. Finally, the image was rotated and the lower-left corner after correction was gained by using rotation formula. Meanwhile, the number region was located by shifting fixed distance from anchor point, which makes the time consumption limited to few microseconds. The experimental results prove good robustness and high time efficiency, so the scheme has a broad application prospects in the system of optical character recognition of banknote.