票据图像中的格线检测与去除

Detection and removal of form lines from bill images

针对票据图像中手写体字符常常与格线交叠的情况,提出了一种直接利用图像灰度信息的格线检测与去除算法。利用字符和格线的边缘信息定位格线并检测字线的交点,然后根据笔画与格线的两种交叠方式(相割与相交)将格线上的像素划分为两个区域:保护区和擦除区,最后动态地选取填充色去除擦除区内的像素。该算法避免了二值化,对806张真实票据中的小写金额域的识别结果比较,显示了该算法的有效性和鲁棒性。

Characters often overlap with form lines, which will greatly affect the performance of recognition system. A line detection and removal algorithm is directly presented based on gray images, while many other algorithms are based on bi-level images. Line positions and cross-points of characters and lines were first detected with edge information. Then pixels interior to a line were divided into two sets： protecting-set and erasing-set by crossing shape analysis. Finally, the gray level of the pixels in erasing-set was set to their local background color. Experiments on 806 real life courtesy amount images demonstrate the efficient of this algorithm, and the recognition rate is up to 91.4%.