发票印刷体数字识别方法的研究

Research on Recognition Method on Invoice Printing Number

在发票图片的采集过程中,由于拍摄不当获取到的图片存在倾斜;受采集环境的影响,采集到的发票图片表面有光照不均匀,部分区域过亮或过暗,不利于数字的定位与识别。针对此类问题,在预处理阶段,首先应用霍夫变换法检测发票图片中的横线并计算其倾斜角,通过旋转对倾斜发票图片进行矫正;其次,对发票图像进行预处理操作,减弱光照以及噪声的影响;接着研究了普通发票版面特征以及数字分布位置,提出了一种基于投影法的定位方法,准确定位出数字区域;最终选用基于数字结构特征的方法判别数字。实验结果表明,该算法识别速度快、精度高。

In the process of invoice collection,some pictures may have the tilt problem due to inappropriate photograph method. In addition,non-uniform illumination may happen as a result of the photograph environment,which adds to the difficulty of location and identification of invoice numbers. In viewof these problems,Hough transform has been applied firstly in the preprocessing section to detect and calculate the tilt angle of the picture,then rotating to correct invoices. Secondly,preprocessing has been arranged for the invoices to eliminate the effect of noise. Thirdly,a method of location based on projection is proposed by the research on the layout of common invoices,which can identify the number area accurately. Finally,the algorithm based on digital structure feature is adopted to identify. The experiment shows that this algorithm has high identification velocity and precision rate.