多特征融合决策的发票印章识别  

Invoice seal identification based on multi-feature fusion decision

作　　者：欧阳欢 范大昭[1] 李东子 OUYANG Huan;FAN Da-zhao;LI Dong-zi(School of Geospatial Information,Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学地理空间信息学院,河南郑州450001

出　　处：《计算机工程与设计》2018年第9期2842-2847,共6页Computer Engineering and Design

摘　　要：识别发票专用章是提取发票信息的关键步骤之一,为此提出一种多特征融合决策的印章识别算法。利用最大对称环绕(MSS)算法检测图像中的感兴趣区域;依据印章的形态特征剔除无关目标,通过迭代自组织数据分析(ISODATA)算法对图像中的剩余目标进行聚类分析;提取各类别外轮廓上的样本点进行随机最小二乘椭圆拟合,实现印章的识别与扶正。实验结果表明,所提算法的识别准确率高,抗噪性好,对印章轮廓残缺或背景复杂的图像均能得到满意的识别结果。Identifying Chinese invoice seal is one of the key steps in invoice information extracting.Based on multi-feature fusion decision,an invoice seal identification method was proposed.The maximum symmetric surround(MSS)algorithm was employed to separate the region of interest from the background.Irrelevant targets were eliminated according to the morphological characteristics of the seal,and the remaining targets in the image were clustered using iterative self-organizing data analysis algorithm(ISODATA).Sample points on each cluster outline were extracted and used for stochastic least square ellipse fitting and invoice seal rotating and identification.Extensive experiments demonstrate that the proposed method has great identification correctness and noise resistance.For images with incomplete seal contour or complex background,the method can also achieve desirable performance.

关 键 词：印章识别 显著性检测 ISODATA聚类 随机最小二乘 椭圆拟合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]