基于图像配准的包裹表单内容提取

发布时间：2018-11-15 11:12

【摘要】：随着电子商务的发展,网络交易的数量急剧增加,导致大量的包裹需要收发、分拣和运输。由于包裹数量巨大,利用机器进行自动分拣的需求强烈。机器自动分拣过程中,图像采集装置获取运动包裹的图像,采用图像处理和识别技术提取包裹图像上的表单,并识别其中条码、收件人地址、收件人手机号码等信息,为自动分拣提供依据,从包裹图像提取表单内容是技术难点之一。本文提出了一种基于图像配准的包裹表单内容提取方法,用于提取表单中填写的内容,即去掉表单中的表格、线条、商标、预先打印的“收件人地址”字样等格式化的内容并将表单图像做归一化处理,从而便于识别表单中填写的内容。本文的主要工作包括：一、提出一种通过最大团综合图像块间几何关系与SIFT特征描述子间相似性的方式来匹配控制点对的方法。在表单图像存在分辨率、亮度、方向、旋转角度、尺度等方面的差异时,能够求得较为准确的变换模型；二、提出一种图像相减操作,在相减前先对两幅图像进行光照调整以应对图像采集时的光照差异,在相减时为对应像素计算一定范围内最恰当的灰度值,以适应非完全精确配准的情况,在相减后根据差异图像确定模板图像,最终根据模板图像与输入的表单图像确定表单中填写的内容,取得了较好的效果。本文方法在SRI-PARCEL1数据集上进行了测试。SRI-PARCEL1数据集中的图像为中国邮政实际使用中的分拣机采集的包裹表单图像,本文提出的方法在该数据集上取得了较好的效果。
[Abstract]:With the development of electronic commerce, the number of network transactions increases rapidly, which results in a large number of packages need to be sent and sent, sorting and transportation. Because of the large number of packages, there is a strong demand for automatic sorting by machines. In the process of automatic sorting, the image acquisition device acquires the image of the moving package, uses the image processing and recognition technology to extract the form on the package image, and recognizes the bar code, the address of the recipient, the mobile phone number of the recipient, and so on. For automatic sorting, extracting form content from package image is one of the technical difficulties. This paper presents a method for extracting the contents of a package form based on image registration, which is used to extract the contents filled in the form, that is, to remove the forms, lines and trademarks from the form. Preprint the words "addressee address" and normalize the image of the form so as to identify the contents filled in the form. The main work of this paper is as follows: first, a method of matching control point pairs by using the geometric relations between the blocks of the maximum cluster synthesis and the SIFT features to describe the similarity between the sub-subsets is proposed. When there are differences in resolution, brightness, direction, rotation angle, scale and so on in the form image, a more accurate transformation model can be obtained. Secondly, an image subtraction operation is proposed, which adjusts the illumination of the two images before subtraction to deal with the illumination difference in image acquisition, and calculates the most appropriate gray value in a certain range for the corresponding pixels when subtracting. In order to adapt to the incomplete and accurate registration, the template image is determined according to the difference image after subtraction, and the content filled in the form is finally determined according to the template image and the input form image, and good results are obtained. The method of this paper is tested on the SRI-PARCEL1 dataset. The image of the SRI-PARCEL1 dataset is the parcel form image collected by the sorting machine in the practical use of China Post. The method proposed in this paper has achieved good results on the data set.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】