面向海关税收监管的彩码检测与随附单据处理方法研究

发布时间：2018-09-12 12:51

【摘要】：随着国际贸易和跨境电子商务的快速发展,海关税收监管和风险决策需要考虑越来越多的数据来源,在传统结构化数据的基础之上,还需要进一步提取和分析非结构化数据。图像文档是海关非结构化数据的一个重要来源,从中提取有用信息,对海关业务开展具有重要意义。本文主要对商品条码照片、报关随附单据两种图像文档进行了着重分析。针对商品条码的检测和识别问题,本文首先提出了一种快速的彩色二维码(彩码)检测方法。该方法将彩码原型视为一组4或8种颜色方块组成的矩形区域。我们首先利用一种基于颜色差异的交叉点检测算子检测彩码图像中存在的交叉点,同时利用二值化和连通区域分析方法产生彩码候选区域。候选区域包含的交叉点大于一定数量,且满足一定的形状特征条件,则被认为是彩码区域。对于在自建的约600张彩码照片数据集上,该方法在取得较高的准确率基础上同时保证了高效率和健壮性。针对文中提出的彩码原型,本文同时提出了其对应的规格推断方法。该方法首先对检测到的彩码区域进行投影变换,然后提取边缘信息,通过分析边缘之间的距离关系最终得到彩码的规格及对应的颜色值矩阵。该方法对平整表面上较清晰彩码图像具有较好的效果。针对报关随附单据的预处理问题,本文改进了一种基于一维卷积的扫描单据线段检测方法。该方法首先使用一个环形中值滤波器消除图像中的块状前景区域,然后分别在水平和垂直方向上进行一维卷积并归一化。继续对得到的卷积图像进行二值化处理、连通区域分析和特征提取,最后使用SVM进行训练,并对检测到的线段进行正确性判断。该方法在55张真实单据图像上同时取得了较高的准确率和召回率。
[Abstract]:With the rapid development of international trade and cross-border e-commerce, customs tax supervision and risk decision-making need to consider more and more data sources. On the basis of traditional structured data, it is necessary to extract and analyze unstructured data. Information is of great significance to the development of customs business.This paper mainly analyzes the barcode photos and the accompanying documents of customs declaration.Aiming at the detection and recognition of barcode,a fast color two-dimensional code(color code) detection method is proposed in this paper.The method regards the color code prototype as a group of four or eight kinds. We first detect the intersections in color coded images by using a crosspoint detection operator based on color difference, and then generate candidate regions by using binarization and connected region analysis. For the self-built 600 color-coded photo datasets, this method guarantees high efficiency and robustness on the basis of high accuracy. For the color-coded prototype proposed in this paper, the corresponding specification inference method is also proposed. Shadow transform is used to extract the edge information, and the size of the color code and the corresponding color value matrix are obtained by analyzing the distance relationship between the edges. This method has a good effect on the clear color code image on the flat surface. Segment detection method. Firstly, a ring median filter is used to eliminate the block foreground region in the image, and then one-dimensional convolution and normalization are carried out in horizontal and vertical directions respectively. The correctness of the line segment is judged. The method achieves high accuracy and recall rate simultaneously on 55 real document images.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】