WO2019071476A1 - 一种基于智能终端的快递信息录入方法及录入系统 - Google Patents

一种基于智能终端的快递信息录入方法及录入系统 Download PDF

Info

Publication number
WO2019071476A1
WO2019071476A1 PCT/CN2017/105743 CN2017105743W WO2019071476A1 WO 2019071476 A1 WO2019071476 A1 WO 2019071476A1 CN 2017105743 W CN2017105743 W CN 2017105743W WO 2019071476 A1 WO2019071476 A1 WO 2019071476A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
image
character
module
express
Prior art date
Application number
PCT/CN2017/105743
Other languages
English (en)
French (fr)
Inventor
梁少勃
Original Assignee
深圳传音通讯有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳传音通讯有限公司 filed Critical 深圳传音通讯有限公司
Priority to PCT/CN2017/105743 priority Critical patent/WO2019071476A1/zh
Priority to CN201780095880.7A priority patent/CN111213157A/zh
Publication of WO2019071476A1 publication Critical patent/WO2019071476A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/10—Image acquisition

Definitions

  • the invention relates to the field of intelligent terminals, in particular to a method and a system for entering a delivery information based on a smart terminal.
  • the types of goods can relate to people's daily life. Every aspect of it. The buyer only needs to place an order through the client product of the e-commerce platform (the website or the app in the mobile terminal), and then can enter a series of processes such as seller delivery, logistics company distribution, etc., and finally the buyer does not leave the house. You can receive the goods you ordered. Among them, the filling of the address of the sender/sender and the delivery of the goods are a very important part. Once the error occurs, the buyer or the seller may be damaged.
  • the present invention provides a smart terminal-based express information entry method and an entry system, which realizes automatic entry of express information without manual or manual input, improves work efficiency, and reduces processing time of a single express; meanwhile, automated Information entry reduces the incidence of errors caused by manual input.
  • an object of the present invention is to provide a smart terminal-based express information entry method and an entry system.
  • a smart terminal-based express information entry method including the following steps:
  • the courier information is entered into the corresponding location on the mailing page.
  • the step of acquiring an information image containing the delivery information comprises:
  • the step of preprocessing the information image comprises:
  • the inclination correction is performed based on the edge of the information image or the direction of the line of the character.
  • the text recognition is performed on each area of the information image after the area division, and the step of extracting the delivery information includes:
  • the express information input method further includes:
  • a smart terminal-based express information entry system includes: a page entry module, an image acquisition module, an image pre-processing module, an image segmentation module, and an information recognition. Other modules and information entry modules;
  • the page enters a module and enters a mailing page of the smart terminal
  • the image acquisition module is in communication with the page entry module to obtain an information image containing the delivery information
  • the image preprocessing module is communicatively coupled to the image acquisition module to preprocess the information image
  • the image segmentation module is communicably connected to the image preprocessing module, and performs layout analysis and region segmentation on the preprocessed information image;
  • the information identification module is communicably connected to the image segmentation module, and performs character recognition on each region of the information image after the segmentation, and extracts the courier information;
  • the information entry module is in communication with the page entry module and the information recognition module, and records the delivery information into a corresponding location of the mailing page.
  • the image acquisition module opens a camera of the smart terminal, and captures and acquires an information image containing the delivery information
  • the image acquisition module calls an album application of the smart terminal to acquire an information image including the delivery information.
  • the image preprocessing module comprises: an image preprocessing unit and a tilt correction unit;
  • the image preprocessing unit performs binarization processing or gray level processing on the information image
  • the inclination correcting unit performs tilt correction based on the edge of the information image or the line direction of the character.
  • the information identification module includes: a character separation unit, a feature extraction unit, and a character matching unit;
  • the character separating unit cuts the information image into image lines, and separates a single character from the image line;
  • the feature extraction unit is communicatively coupled to the character separation unit to extract statistical features or structural features from the single character, including refinement and normalization;
  • the character matching unit is communicably connected to the feature extraction unit, and finds a character class with the highest similarity to the single character from the learned feature library.
  • the express information entry system further includes an information storage module, and is communicably connected to the information identification module to store the express delivery information.
  • the express information input method and the input system provided by the invention realize automatic entry of express information without manual or manual input, thereby improving work efficiency and reducing processing time of a single express; meanwhile, automated information Entry reduces the incidence of errors caused by manual input.
  • FIG. 1 is a schematic flow chart of a method for entering a delivery information according to a preferred embodiment of the present invention
  • FIG. 2 is a schematic flow chart of an image pre-processing step of the express information input method of FIG. 1;
  • FIG. 3 is a schematic flow chart of a step of identifying a delivery information of the express information input method of FIG. 1;
  • FIG. 4 is a block diagram showing the structure of a courier information entry system in accordance with a preferred embodiment of the present invention.
  • connection should be understood broadly, and may be, for example, a mechanical connection or an electrical connection, or may be internal to the two elements, or may be The direct connection may also be indirectly connected through an intermediate medium.
  • connection should be understood broadly, and may be, for example, a mechanical connection or an electrical connection, or may be internal to the two elements, or may be The direct connection may also be indirectly connected through an intermediate medium.
  • specific meanings of the above terms may be understood according to specific situations.
  • module or "unit” for indicating an element is merely an explanation for facilitating the present invention, and does not have a specific meaning per se. Therefore, “module” and “unit” can be used in combination.
  • the information input method and the information input system of the present invention can be applied to smart terminals, and the smart terminals can be in various forms.
  • the smart terminal described in the present invention can include, for example, a mobile phone, a smart phone, a notebook computer, a PDA (personal number).
  • Mobile terminals such as assistants, PADs (tablets), PMPs (portable multimedia players), navigation devices, smart watches, and the like, and fixed terminals such as digital TVs, desktop computers, and the like.
  • the terminal The present invention will be described with respect to a mobile terminal and assuming that the mobile terminal is a smart phone.
  • a smart terminal-based express information input method includes the following steps:
  • S400 performing layout analysis and region segmentation on the preprocessed information image
  • S500 performing character recognition on each area of the information image after the area division, and extracting the express delivery information
  • S200 Obtain an information image containing the delivery information; in a preferred embodiment, S200: the step of acquiring an information image containing the delivery information includes:
  • the method for obtaining the information image may be various, and the image may be captured on the spot by calling the smart terminal camera or randomly selected from the local or cloud album of the smart terminal.
  • the delivery information includes but is not limited to: the recipient/sender name, phone number, region, and detailed address.
  • the step of pre-processing the information image includes:
  • S310 Perform binarization processing or gray level processing on the information image.
  • S320 Perform tilt correction based on the edge of the information image or the line direction of the character.
  • the problem of uneven illumination is an important part of the identification and extraction of express information in information images.
  • Highlights and shadows are the two main factors that affect image uniformity.
  • the highlight area of the information image can automatically correct the highlight area in the image by the color shift rate of each component in the R, G, B space of the image.
  • the Retinex method can be used to correct the shadow of an optical image and enhance the image.
  • the highlight region correction can use the pixel and spatial distribution features of the information image to detect the normal illumination region, and then automatically correct the non-uniform illumination in the image using the color shift rate of each component.
  • Weighted average method On the basis of the average method, according to the principle of sensitization of the human eye, the coefficients of each channel are converted into different values, thereby obtaining a result closer to the human eye observation. Its mathematical expression is as follows:
  • the maximum method usually obtains a high-intensity grayscale image, while the averaging method tends to blur the edge details of the image to obtain a soft-edged image, and the weighted average method can be retained by adjusting the values of k 0 , k 1 , and k 2 .
  • An image of multiple image high frequency information ie, edge information). Therefore, the present invention uses a weighted averaging method to gradation of a non-uniformly corrected information image.
  • the information image after non-uniform correction and image graying reduces the influence of non-uniform illumination and image color on text recognition.
  • the express information in the information image is mainly text, and the main feature of the text is structural features.
  • the structural feature is a feature quantity independent of the gray level of the image, so the information image also needs to be binarized, that is, the gray image is reduced to a (0, 1) image. Binarization is usually expressed as:
  • the binarization of grayscale images requires effective discrimination of character pixels from the image.
  • the binarization requirement of the traditional printed font image requires on the one hand to maintain the structural features of the original characters as much as possible, and on the other hand, it is required that there is no blank between the binarized characters.
  • the background of traditional printed characters is relatively simple and smooth, and the background and characters can be distinguished by a simple global threshold.
  • the layout analysis of the image includes the tilt correction of the image and the segmentation of the image rows and columns.
  • the image tilt correction method mainly includes a projection map method, a straight line fitting method, and a Hough variation method.
  • the present invention can be adjusted by paying attention to the angle at which an image is placed, etc., and the image tilting can be ignored for the sake of simplifying the processing. Therefore, the main content of the information image layout analysis of the present invention is the row and column division of the information image.
  • Layout analysis is one of the key technologies for optical image based recognition systems. Through the analysis of the information image, it is important to determine the location of the sender/sender's name, gender, address, telephone number and other express information in the information image.
  • the layout of the information image of general delivery information is relatively fixed, it is more complicated. First of all, there must be some differences in the layout of information images, and it is necessary to choose a method with better fitness. Secondly, the font size of the same line in the information image may be different. For example, in the first line name information, the name font is smaller than the font of "name”, which brings great trouble to the line extraction in the layout analysis. In the image again, there is a phenomenon in which the numbers and the Chinese characters coexist, which brings great difficulty to the division of the character blocks.
  • the layout analysis method of the image includes a top-up method and a bottom-up method.
  • the method of the top-down method starts with the macroscopic direction of the image, and gradually analyzes the image into different modules by analyzing the global features of the image. By dividing it over and over again, the image is finally divided into individual structural elements.
  • the start position and the end position of the character line are first determined, and the position of the single character is determined in each line.
  • the top-down method is effective for character recognition with relatively fixed layout. However, for complex layouts, it is difficult to accurately segment the structural elements such as characters, tables, and images in the image because a large number of image details are ignored. .
  • the bottom-up approach is based on the basic structural elements of the image, and is gradually merged into characters, images, or tables by structural analysis of local elements; and then by analyzing the positional relationship between characters, tables, or images, the image layout is obtained. Row and column information. The structure in the image is analyzed column by column, thereby completing the extraction of the entire layout information. From bottom-down method design to a large number of iterative operations, the calculation process is complex, the calculation speed is slow, and it is rarely applied in practice. At present, a large number of optical character-based text recognition methods mainly combine the top-down method and the bottom-up method to balance the recognition speed and performance.
  • S500 performing character recognition on each area of the information image after the area division, and the step of extracting the delivery information includes:
  • S520 Extract statistical features or structural features from the single character, including refinement and normalization
  • the courier information in the optical image needs to be efficiently classified and accurate to each character.
  • the image to be identified itself contains a large amount of information. Taking a 64 ⁇ 64 image as an example, if the image is directly matched with the image to be identified, it will constitute a feature vector of 4096 dimensions, which greatly increases the search space of the recognizer.
  • the storage space of the character template required to directly match the character image will be very large. At present, the recorded Chinese characters are about 90,000 words.
  • the character feature extraction is mainly to complete the features that are substantially different between the characters extracted from the original optical image data.
  • the extraction of character features needs to follow some of the following rules:
  • Distinction There is a big difference between different characters. For example, there is a big difference between Chinese characters and numbers, and different differences between Chinese characters are also required. Different Chinese characters should have a large distance in the feature space. Such a feature can distinguish different characters in the case of image noise or other interference.
  • the above feature requirements are met and at the same time, the number of features is required to be as small as possible.
  • As few feature components as possible can ensure the effective information input space of the recognizer on the one hand, and reduce the storage space required for matching the template on the other hand, and at the same time reduce the search space of the recognizer and speed up the recognition process.
  • Character normalization includes position normalization and size normalization.
  • Character position normalization has two methods: center of gravity normalization and frame normalization.
  • Gravity Normalization Firstly, the position of the center of gravity of the printed character image is calculated. After calculating the position of the center of gravity of the image, the center of gravity of the character is moved to the center of the image, and the position of the character image is normalized after the segmentation.
  • Normalization of the frame Firstly calculate the upper, lower, left and right borders of the character image. After calculating the center position of the image, move the center position of the character to the center of the image to complete the normalization of the character image.
  • the method of normalizing according to the size of the image frame is simple in operation and small in calculation amount.
  • Another method of image size normalization needs to consider the distribution characteristics of the image.
  • One of the simpler methods of image distribution is the distribution variance of the image.
  • the structural features of Chinese characters include stroke features and component features.
  • the stroke characteristics of Chinese characters include the type of stroke and the length of the stroke.
  • the types of strokes mainly include: horizontal, vertical, ⁇ , ⁇ , fold, and hook.
  • the statistics show the proportion of the six strokes in the Chinese characters.
  • the point on the stroke can be used as the feature point of the Chinese character, and in order to further distinguish the structure, the important point on the character background can also be used as the feature point of the character.
  • the points on the stroke and the points in the background of the characters can together form an important feature vector for Chinese character recognition.
  • Feature points in strokes include endpoints, vertices, and ambiguities.
  • the structure of the stroke relationship of Chinese characters is relatively simple, and the relationship between components is very complicated. Different combinations of parts can be made in different numbers and positions. In terms of the number of components, there are single characters, double-word characters, three-body characters, four-body characters, and the like. According to the spatial positional relationship of the components, there are independent fonts, left and right fonts, upper and lower fonts, and bound fonts.
  • the four-direction decomposition of Chinese characters is based on the structural features of Chinese characters, and the Chinese character strokes are decomposed from the four directions of horizontal, vertical, ⁇ , and ⁇ .
  • the simplest four-direction decomposition judges the direction of the Chinese stroke by judging the points in the eight fields of the pixel. There are “and” methods and “or” methods for judging the stroke direction of Chinese characters in eight fields.
  • the above method assumes that the stroke of a Chinese character is a single pixel point. Therefore, the input Chinese character image is required to pass the pen. An image that is thinned or extracted by the stroke skeleton. However, in actual use, the refinement of the Chinese character stroke or the skeleton extraction is prone to the loss of the stroke for the character image with a thin stroke itself, thereby reducing the recognition probability of the character.
  • Statistical features of printed characters based on optical images include both global features and local features. Different from the structural features of characters, the statistical features of characters are obtained from binarized images, and some of them are obtained directly from grayscale images by corresponding transformations.
  • the global feature of a character image essentially treats the character image as a normal image, and the character is only an object with a certain feature. Therefore, the global feature extraction method of the character image is similar to the feature extraction method of the general image.
  • the global feature extraction methods for character images mainly include the following:
  • Variation domain feature component The binarized character image is transformed into other feature spaces, and the coefficients of the corresponding vector in the feature space are taken as features.
  • Commonly used transforms include 2-D Fourior Transformation, Hadam Transformation, Hough Transformation, and the like.
  • the two-dimensional Fourier transform transforms the image information in the spatial domain into a two-dimensional frequency domain, and transforms the spatial position change with greater correlation into the frequency domain composed of the frequency components with normalized orthogonality, thereby obtaining the target. (ie) different characteristics between characters.
  • Hada code transform is a commonly used feature transform in remote sensing images. It uses the symmetric orthogonal Had code matrix to realize the transformation of image spatial relationship to multi-spectral domain, which can extract target features and classify and identify remote sensing images.
  • Hough transform is one of the basic methods for recognizing a target of a specific shape from an optical image
  • the Hough transform for object feature extraction after expansion is mainly to transform an image from a spatial domain into a feature space composed of different basic shapes. , using the coefficients of different geometric shapes to form the characteristics of the target.
  • Invariant moment is an important method for target detection and recognition in optical image processing.
  • the center moment and the origin moment of the image can distinguish the geometry information of the target projected on the imaging plane, but the geometry of the projection surface does not have scale, rotation or affine invariance.
  • Global projection feature The image is respectively projected to several reference directions, and only the stroke perpendicular to the reference direction is projected to the reference direction. Compared with the stroke extraction method based on structural features, the method is simple in operation and can realize fast extraction.
  • the global projection feature can reflect the complexity of the entire Chinese character to a certain extent, the main direction of the stroke and the possible connection relationship between the strokes. To simplify the calculation, the projections in four directions are usually taken, that is, the horizontal direction, the vertical direction, the positive 45 degree direction, and the negative 45 degree direction.
  • the background part and the stroke of the Chinese character can also be used as a global feature of the Chinese character image.
  • blank points (non-stroke points) located on two diagonal lines of the image are selected to count the stroke density of the points in each direction of the character as the global background feature of the image.
  • the local features of Chinese characters need to first divide the image into different local regions and count them in different regions.
  • the features of the image may be one or a combination of the aforementioned global features.
  • the key points of local feature extraction are in the method of partitioning with local regions.
  • Local area division methods include grid method (including fixed grid method and elastic grid method), cell division method, direction line division method, four-corner feature division method and Gabor division method.
  • the single character image obtained by image preprocessing is normalized, and the character feature vector after character feature extraction needs to be sent to the Chinese character classifier to complete the extraction of the delivery information.
  • the unknown sample that the classifier is about to input is divided into different categories, and the tasks of the sample to be identified and the categories to which the samples belong are one-to-one.
  • Image classifiers mainly include Euclidean distance classifiers, neural network classifiers, support vector machine classifiers, and genetic algorithm classifiers.
  • the Euclidean distance classifier is the simplest and most intuitive classification method, and the distance of the points in the high-dimensional space is used as the main basis for the sample similarity measure. The smaller the distance value, the higher the degree of similarity between the samples to be tested.
  • Support vector machine classifier neural network has good curve fitting ability and target classification ability, and has a large number of applications in target recognition and detection.
  • the shortcomings of neural networks are also very obvious.
  • the current structure of the neural network has no reliable rules, so the convergence speed of the algorithm is very slow, and the initial value selection of the network has a great influence on the performance of the algorithm, and the algorithm easily converges to a minimum value.
  • the support vector machine uses a special type of operation plane. This hyperplane is called the optimal classification hyperplane.
  • the present invention also discloses a smart terminal-based express information input system 100.
  • the express information input system 100 includes a page entry module 11, an image acquisition module 12, an image pre-processing module 13, and an image segmentation module 14. , information identification module 15, information entry module 16;
  • the page enters the module 11 and enters a mailing page of the smart terminal
  • the image obtaining module 12 is communicably connected to the page entering module 11 to obtain an information image containing the delivery information.
  • the image preprocessing module 13 is communicatively coupled to the image acquisition module 12 to preprocess the information image;
  • the image segmentation module 14 is communicatively coupled to the image preprocessing module 13 to perform layout analysis and region segmentation on the preprocessed information image;
  • the information identification module 15 is communicably connected to the image segmentation module 14 to perform character recognition on each region of the information image after the region division, and extract the courier information;
  • the information entry module 16 is communicatively coupled to the page entry module 11 and the information recognition module 15 to record the express delivery information into a corresponding location on the mailing page.
  • the image acquisition module 12 turns on the camera of the smart terminal, and captures and acquires an information image containing the delivery information
  • the image obtaining module 12 invokes an album application of the smart terminal to acquire an information image including the delivery information.
  • the image preprocessing module 13 includes: an image preprocessing unit and a tilt correction unit;
  • the image preprocessing unit performs binarization processing or gray level processing on the information image
  • the inclination correcting unit performs tilt correction based on the edge of the information image or the line direction of the character.
  • the information identification module 15 includes: a character separation unit, a feature extraction unit, and a character matching unit;
  • the character separating unit cuts the information image into image lines, and separates a single character from the image line;
  • the feature extraction unit is communicatively coupled to the character separation unit to extract statistical features or structural features from the single character, including refinement and normalization;
  • the character matching unit is communicably connected to the feature extraction unit, and finds a character class with the highest similarity to the single character from the learned feature library.
  • the express information entry system 100 further includes an information storage module 17 communicatively coupled to the information identification module 16 for storing the express delivery information.

Abstract

本发明提供了一种基于智能终端的快递信息录入方法及录入系统，其中，录入方法包括以下步骤：进入所述智能终端的寄件页面；获取一包含快递信息的信息图像；对所述信息图像进行预处理；对预处理后的所述信息图像进行版面分析及区域分割；对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息；将所述快递信息录入所述寄件页面的相应位置。本发明实现了快递信息的自动录入，无需手写或人工输入，提高了工作效率，降低了单个快递的处理时间；同时，自动化的信息录入，降低了由手动输入导致的错误发生率。

Description

一种基于智能终端的快递信息录入方法及录入系统 技术领域
本发明涉及智能终端领域，尤其涉及一种基于智能终端的快递信息录入方法及录入系统。
背景技术
自首款智能手机问世以来，智能手机制造商就在不断的改进产品设计，越来越多无法想象到的功能来到我们身边，随着智能手机功能的丰富以及移动互联网的快速发展，智能手机已然取代了我们身边很多常用的电子设备，改变着我们的生活方式以及周边的行业。
随着电子商务交易平台的不断完善，以及传统通信、移动通信等技术的快速发展，越来越多的人们通过网上购物的方式来获取自己所需的商品，商品的种类可以涉及到人们日常生活的方方面面。买家只需要通过电商平台的客户端产品(网站或者移动终端中的应App)进行下单，之后就可以进入卖家发货、物流公司配送等一系列流程，最终使得买家足不出户就可以接收到自己订购的货品。其中，收/寄件人地址的填写、货品的配送是非常关键的一环，一旦发生送错等现象，可能会给买家或者卖家造成损失。当前，较为传统的寄件方式为采用纸质的快递货单，需要手动填写收件人、寄件人地址等个人信息，一方面，手写效率低下、严重降低工作效率，且经常需要重复毫无意义的工作，另一方面，快递信息的反馈需要记录快递单号。随后兴起的电子寄件，通过扫描二维码等方式，相比原始寄件更加快速，但是，收件人、寄件人地址等个人信息仍然需要手动输入，信息的录入仍然相当繁琐。
当前，基于扫描图像的信息录入方法已经被广泛应用。在书报刊行业和其它对纸质阅读材料依赖较大的行业，采用扫描仪扫描书本，并将扫描得到的图片内容组合成保存完整的书本，之后将该书本文件转化为PDF文件，这种方式可以在较短的时间内将大量的纸质内容转化为电子文件保存在物理存储介质中。基于扫描图像的信息录入拥有录入时间短、录入信息量大、录入信息损失小等特点。
鉴于当前电商的快速发展和快递行业的业务量持续增大，依靠快递从业人员手工录入快递信息的方式将会越来越难以适应这个行业大数量、短时间的需求。因此，我们迫切需要一种使用方便快捷、信息采集录入准确度较高、性能稳定、适用性较强的快递信 息录入方法及录入系统，以解决现有快递行业需要人工录入快递信息的缺点，提高工作效率。
因此，本发明提供了一种基于智能终端的快递信息录入方法及录入系统，实现快递信息进行自动录入，无需手写或人工输入，提高了工作效率，降低了单个快递的处理时间；同时，自动化的信息录入，降低了由手动输入导致的错误发生率。
发明内容
为了克服上述技术缺陷，本发明的目的在于提供一种基于智能终端的快递信息录入方法及录入系统。
本发明的一方面，公开了一种基于智能终端的快递信息录入方法，包括以下步骤：
进入所述智能终端的寄件页面；
获取一包含快递信息的信息图像；
对所述信息图像进行预处理；
对预处理后的所述信息图像进行版面分析及区域分割；
对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息；
将所述快递信息录入所述寄件页面的相应位置。
优选地，获取一包含快递信息的信息图像的步骤包括：
开启所述智能终端的摄像头，拍摄并获取一包含快递信息的信息图像；
和/或
调用所述智能终端的相册应用，获取一包含快递信息的信息图像。
优选地，对所述信息图像进行预处理的步骤包括：
将所述信息图像进行二值化处理或灰度级别处理；
以所述信息图像的边沿或文字的行向为基准进行倾斜度校正。
优选地，对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息的步骤包括：
将所述信息图像切割为图像行，从所述图像行中分离出单个字符；
从所述单个字符上提取统计特征或结构特征，包括细化和归一化；
从学习得到的特征库中找到与所述单个字符相似度最高的字符类。
优选地，所述快递信息录入方法还包括：
存储所述快递信息。
本发明另一方面，公开了一种基于智能终端的快递信息录入系统，所述快递信息录入系统包括：页面进入模块、图像获取模块、图像预处理模块、图像分割模块、信息识 别模块、信息录入模块；
所述页面进入模块，进入所述智能终端的寄件页面；
所述图像获取模块，与所述页面进入模块通信连接，获取一包含快递信息的信息图像；
所述图像预处理模块，与所述图像获取模块通信连接，对所述信息图像进行预处理；
所述图像分割模块，与所述图像预处理模块通信连接，对预处理后的所述信息图像进行版面分析及区域分割；
所述信息识别模块，与所述图像分割模块通信连接，对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息；
所述信息录入模块，与所述页面进入模块、信息识别模块通信连接，将所述快递信息录入所述寄件页面的相应位置。
优选地，所述图像获取模块，开启所述智能终端的摄像头，拍摄并获取一包含快递信息的信息图像；
和/或
所述图像获取模块，调用所述智能终端的相册应用，获取一包含快递信息的信息图像。
优选地，所述图像预处理模块包括：图像预处理单元、倾斜度校正单元；
所述图像预处理单元，将所述信息图像进行二值化处理或灰度级别处理；
所述倾斜度校正单元，以所述信息图像的边沿或文字的行向为基准进行倾斜度校正。
优选地，所述信息识别模块包括：字符分离单元、特征提取单元、字符匹配单元；
所述字符分离单元，将所述信息图像切割为图像行，从所述图像行中分离出单个字符；
所述特征提取单元，与所述字符分离单元通信连接，从所述单个字符上提取统计特征或结构特征，包括细化和归一化；
所述字符匹配单元，与所述特征提取单元通信连接，从学习得到的特征库中找到与所述单个字符相似度最高的字符类。
优选地，所述快递信息录入系统还包括信息存储模块，与所述信息识别模块通信连接，存储所述快递信息。
采用了上述技术方案后，与现有技术相比，具有以下有益效果：
1.本发明提供的快递信息录入方法及录入系统，实现快递信息进行自动录入，无需手写或人工输入，提高了工作效率，降低了单个快递的处理时间；同时，自动化的信息 录入，降低了由手动输入导致的错误发生率。
附图说明
图1为符合本发明一优选实施例的快递信息录入方法的流程示意图；
图2为图1的快递信息录入方法的图像预处理步骤的流程示意图；
图3为图1的快递信息录入方法的快递信息识别步骤的流程示意图；
图4为符合本发明一优选实施例的快递信息录入系统的结构示意图。
附图标记：
100-快递信息录入系统；11-页面进入模块；12-图像获取模块；13-图像预处理模块；14-图像分割模块；15-信息识别模块；16-信息录入模块；17-信息存储模块。
具体实施方式
以下结合附图与具体实施例进一步阐述本发明的优点。
这里将详细地对示例性实施例进行说明，其示例表示在附图中。下面的描述涉及附图时，除非另有表示，不同附图中的相同数字表示相同或相似的要素。以下示例性实施例中所描述的实施方式并不代表与本发明相一致的所有实施方式。相反，它们仅是与如所附权利要求书中所详述的、本发明的一些方面相一致的装置和方法的例子。
在本发明使用的术语是仅仅出于描述特定实施例的目的，而非旨在限制本公开。在本发明和所附权利要求书中所使用的单数形式的“一”、“所述”、“该”等也旨在包括多数形式，除非上下文清楚地表示其他含义。还应当理解，本文中使用的术语“和/或”是指并包含一个或多个相关联的列出项目的任何或所有可能组合。
在本发明的描述中，除非另有规定和限定，需要说明的是，术语“连接”应做广义理解，例如，可以是机械连接或电连接，也可以是两个元件内部的连通，可以是直接相连，也可以通过中间媒介间接相连，对于本领域的普通技术人员而言，可以根据具体情况理解上述术语的具体含义。
在后续的描述中，使用用于表示元件的诸如“模块”或“单元”的后缀仅为了有利于本发明的说明，其本身并没有特定的意义。因此，“模块”与“单元”可以混合地使用。
本发明的信息录入方法及信息录入系统，可以应用于智能终端，智能终端可以以各种形式，例如，本发明中描述的智能终端可以包括诸如移动电话、智能电话、笔记本电脑、PDA(个人数字助理)、PAD(平板电脑)、PMP(便携式多媒体播放器)、导航装置、智能手表等的移动终端，以及诸如数字TV、台式计算机等的固定终端。下面，假设终端 是移动终端，并假设该移动终端为智能手机，对本发明进行说明。然而，本领域技术人员将理解的是，除了特别用于移动目的的元件之外，根据本发明的实施方式的构造也能够应用于固定类型的终端。为便于描述，本发明实施例均以智能手机为例进行说明，其它应用场景相互参照即可。
参考图1，为本发明一优选实施例的基于智能终端的快递信息录入方法，包括以下步骤：
S100：进入所述智能终端的寄件页面；
S200：获取一包含快递信息的信息图像；
S300：对所述信息图像进行预处理；
S400：对预处理后的所述信息图像进行版面分析及区域分割；
S500：对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息；
S600：将所述快递信息录入所述寄件页面的相应位置。
具体地：
-S100：进入所述智能终端的寄件页面；
在使用智能终端进行电子寄件时，进入快递寄件页面，添加新增地址，需要输入收/寄件人信息时，显示“拍摄或上传图像，自动识别”的文字或其他提示，提示用户拍摄包含快递信息的信息图像。
-S200：获取一包含快递信息的信息图像；在一优选实施例中，S200：获取一包含快递信息的信息图像的步骤包括：
开启所述智能终端的摄像头，拍摄并获取一包含快递信息的信息图像；
和/或
调用所述智能终端的相册应用，获取一包含快递信息的信息图像。
获取信息图像的方法可以是多种，可以通过截图方式，或者调用智能终端摄像头当场拍摄图像，或者从智能终端的本地或云端相册中任意选取一张图像。
此处，快递信息包括但不限于：收/寄件人名称、电话、所在地区、详细地址等信息。
-S300：对所述信息图像进行预处理；
参考图2，在一优选实施例中，S300：对所述信息图像进行预处理的步骤包括：
S310：将所述信息图像进行二值化处理或灰度级别处理；
S320：以所述信息图像的边沿或文字的行向为基准进行倾斜度校正。
-S400：对预处理后的所述信息图像进行版面分析及区域分割。
对图像的预处理、版面分析及区域分割：
-非均匀光照校正
光照不均匀问题是信息图像中快递信息识别提取的一个重要环节。高光和阴影是影响图像均匀性的两个主要因素。信息图像的高光区域可以通过图像在R，G，B空间中各分量颜色偏移率自动校正图像中的高光区域。Retinex方法可用于光学图像的阴影进行校正，对图像进行增强操作。
现实中图像采集受到不均匀光照的影响，图像可能存在部分高亮度区域。高光区域校正可使用信息图像的像素与空间分布特征，检测正常光照区域，然后利用各个分量的颜色偏移率，自动校正图像中的非均匀光照。
-灰度化
图像灰度化主要有三种方法：
(1)最大值法：即从R，G，B三个通道的灰度值中选取最大的一个作为当前像素点的灰度值。其数学表达式如下：
G＝max(R，G，B)
(2)平均值法：将R，G，B三个通道的灰度值直接相加求得一个平均值作为当前像素点的灰度值。其数学表达式如下：
G＝(R+G+B)/3
(3)加权平均值法：在平均值法的基础上，依据人眼的感光原理，将原来每一个通道的系数变换为不同的值，从而得到更加接近于人眼观测的结果。其数学表达式如下：
G＝k0R+k1G+k2B
其中k0、k1、k2满足如下的关系：
k0+k1+k2＝1
最大值法通常会得到高亮度的灰度图像，而平均法往往会模糊图像的边沿细节，得到边沿柔和的图像，而加权平均法通过调整k0、k1、k2的值可以得到保留较多图像高频信息(即边沿信息)的图像。因此本发明采用加权平均法对经过非均匀校正后的信息图像进行灰度化。
-二值化
经过非均匀校正和图像灰度化的信息图像减小了非均匀光照和图像色彩对文字识别的影响。信息图像中的快递信息主要为文字，文字的主要特征为结构特征。结构特征是独立于图像灰度的一种特征量，因此信息图像还需要经过二值化处理，即将灰度图像简化为(0，1)图像。二值化通常表示为下式：
Figure PCTCN2017105743-appb-000001
灰度图像的二值化要求有效从图像中区分字符像素。传统印刷字体图像的二值化要求一方面要求尽量保持原始字符的结构特征，另一方面要求二值化后的字符之间不能有空白。传统印刷字符的背景比较单一且平滑，背景与字符之间可以利用简单的全局阈值的方法区分。
-版面分析
图像的版面分析包括了图像的倾斜校正和图像行列的分割。图像倾斜校正方法主要有投影图方法、直线拟合方法以及Hough变化方法等。本发明可以通过采集图像时注意摆放的角度等来调整，为简化处理可不考虑图像倾斜的情况。因此本发明信息图像版面分析的主要内容为信息图像的行列分割。版面分析是基于光学图像的识别系统的关键技术之一。通过对信息图像的版面分析，确定收件人/寄件人的姓名、性别、地址、电话号码等快递信息在信息图像中的位置，对准确分析快递信息具有重要意义。虽然，一般快递信息的信息图像的版面比较固定，但是却比较复杂。首先信息图像的排版必然有一些差别，必须选择适应度较好的方法。其次信息图像中同一行字体大小可能有差别，例如第一行姓名信息中，名字字体比“姓名”的字体小，给版面分析中的行提取带来很大的麻烦。再次图像中存在数字和汉字字体共存的现象，给字符块的划分带来了较大的困难。
图像的版面分析方法包括自顶向上的方法和自底向上的方法。自顶向下方法的方法首先从图像的宏观方向着手，通过分析图像的全局特征，逐渐将图像区分为不同的模块。通过一次又一次迭代划分，最后将图像区分为一个个结构元素为止。本发明中，在文字区域的划分过程中，首先确定文字行的起始位置和结束位置，其次在每一个行中确定单个字符的位置。自顶向下的方法对于版面相对固定的字符识别比较有效，但对于版面比较复杂的情况，由于忽略了大量的图像细节，很难对图像中的字符、表格、图像等结构元素进行准确的分割。
自底向上的方法是从图像基本结构元素出发，通过局部元素的结构分析，逐渐合并成一个个字符、图像或者表格；再通过分析字符、表格或者图像之间的位置关系，得到图像版面中的行、列信息。逐行逐列地分析图像中的结构，从而完成对整个版面信息的提取工作。自底向下的方法设计到大量的迭代运算，计算过程复杂，计算速度慢，在实际中应用较少。目前大量的基于光学字符的文字识别方法主要是将自顶向下的方法和自底向上的方法相结合，从而在识别速度和性能两者之间取得平衡.
-S500：对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信 恩：
参考图3，在一优选实施例中，S500：对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息的步骤包括：
S510：将所述信息图像切割为图像行，从所述图像行中分离出单个字符；
S520：从所述单个字符上提取统计特征或结构特征，包括细化和归一化；
S530：从学习得到的特征库中找到与所述单个字符相似度最高的字符类。
对信息图像中字符特征的提取：
要从光学图像中自动识别出快递信息，需要将光学图像中的快递信息有效进行分类处理，并且精确到每一个字符。然而直接使用图像预处理后得到的图像进行匹配是不现实的。首先待识别的图像本身会包含大量的信息，以64×64大小的图像为例，如果直接以待识别的图像进行匹配会构成4096维的特征向量，大大增加了识别器的搜索空间。其次直接以字符图像进行匹配所需要的字符模板的存储空间会非常大。目前在记录的中国汉字大约为9万字左右，虽然常用的汉字只有3500字左右，但是随着人民文化水平的提高，快递信息中尤其是收/寄件人姓名一项中，出现生僻字的概率越来越大，因此如果取字符的原始图像进行存储，即使单一字体的字符存储量也非常大，更不用说不同省份之间字体稍有差别。最后直接使用的图像会受图像传感器采样噪声、图像采集角度等影响，带来较大的匹配误差。因此，在图像预处理后需要对光学图像中的字符特征进行提取，以减小字符识别器的搜索空间。
字符特征提取主要是完成从原始的光学图像数据中提取出字符之间具有本质不同的特征。字符特征的提取需要遵循以下的一些规则：
(1)区别性：即不同的字符之间有较大的区别。例如汉字字符和数字之间需要有较大的区别，汉字字符之间也需要有不同的区别。不同的汉字在特征空间内应该具有较大的距离。这样的特征才能在图像有采样噪声或者其他干扰的情况下对不同的字符进行区分处理。
(2)可靠性：即同一个字符即使在噪声或者旋转、缩放等情况下具有相同或者非常近似的特征向量。
(3)独立性：即要求同一字符的不同特征之间不相关。特征分量之间的独立性保证了单一特征量不变化不影响其他特征量的判断。
在字符特征提取的过程中，在满足以上特征要求和同时，要求特征数量尽量少。尽量少的特征分量能一方面保证识别器有效的信息输入空间，另一方面可以减小匹配模板所需要的存储空间，同时还能减小识别器的搜索空间，加快识别进程。
-字符归一化处理
经过图像整体预处理后分割得到的字符块图像的大小存在较大的差别。图像块大小不同不会影响字符的结构特征，然而汉字笔画的长度也是汉字的一个非常重要的特征。汉字笔划的长度特征与图像的大小成正比关系，因此经过字符切分后获得的字符图像在特征提取之前尚需要进行字符的归一化。字符归一化处理包括位置归一化和大小归一化。
字符位置归一化有重心归一化和边框归一化两种方法。重心归一化首先计算印刷体字符图像的重心的位置，计算得到图像的重心位置后，将字符重心移到图像中心位置，完成切分后字符图像的位置归一化。边框归一化首先计算出字符图像的上、下、左、右边框，计算得到图像的中心位置后，将字符的中心位置移到图像中心，完成字符图像的归一化。
图像大小归一化有两种方式，一种是按照字符图像外框大小进行归一化：即按照图像外框大小将图像放大或者缩小到规定某一个特定的大小。按照图像外框的大小进行归一化操作的方法操作简单、运算量较小。另一种图像大小归一化的方法需要考虑图像的分布特征。图像的分布特征中比较简单的一种方法是图像的分布方差。
-字符的结构特征
汉字字符的结构特征包括笔划特征和部件特征。
汉字的笔划特征包括笔划的类型以及笔画的长度。笔划的类型主要包括：横、竖、撇、捺、折、钩六类。关统计资料显示6种笔划在汉字字符中所占的比例，汉字中横、撇、竖、捺四种基本笔划占了绝大部分，而折、钩可以近似认为是由以上四种基本笔划构成。因此汉字的笔划可以作为识别的一个重要特征。也可以利用横、竖、撇、捺四个方向对字符的结构特征进行统计分析。
笔划上的点可以作为汉字字符的特征点，同时为了对结构进行进一步区分，字符背景上的重要点也可以作为字符的特征点。笔划上的点和字符背景中的点可以共同构成汉字识别的重要特征向量。笔划中的特征点包括端点、折点、歧点等。
汉字的笔划关系结构比较简单，而部件之间关系则非常复杂。部件的数量和位置不同还可以出现不同的组合。从部件的数量上来看有独体字、双体字、三体字、四体字等。按照部件的空间位置关系有独立字体、左右字体、上下字体、包围字体。
汉字的四方向分解即根据汉字的结构特征，从横、竖、撇、捺四个方向对汉字笔划进行分解。最简单的四方向分解通过判断像素点的八领域内的点来判断汉字笔划的方向。八领域的汉字笔划方向的判别方法有“且”方法和“或”方法。
上述方法假设汉字字符的笔划为单像素点。因此要求输入的汉字字符图像为经过笔 划细化或者笔划骨架提取的图像。然而在实际使用中，汉字笔划的细化或者骨架提取对于笔划本身较细的字符图像容易出现笔划丢失的情况，从而降低字符的识别概率。
-字符的统计特征
基于光学图像的印刷体字符的统计特征包括全局特征和局部特征两种。与字符的结构特征不同，字符的统计特征从二值化图像，有部分甚至直接从灰度图像中通过相应的变换获取字符的特征。
字符图像的全局特征本质上只是把字符图像当作一种普通的图像进行处理，字符只是其中具有某一特征的物体。因此字符图像的全局特征提取方法与一般图像的特征提取方法类似。字符图像的全局特征提取方法主要有以下几种：
(1)变化域特征分量：将二值化后的字符图像变换到其他特征空间，将特征空间中对应向量的系数作为特征。常用的变换有二维傅立叶变换(2-D Fourior Transformation)、哈达码变换(Hadam Transformation)、霍夫变换(Hough Transformation)等。二维傅立叶变换即将空间域的图像信息变换到二维的频率域，将具有较大相关性的空间位置变化变换到由具有归一化正交性的频率分量构成的频率域中，从而获得目标(即)字符之间不同的特征。哈达码变换是遥感图像中一种常用的特征变换，它利用对称正交的哈达码矩阵实现图像空间关系到多光谱域的变换，达到提取目标特征、对遥感图像进行分类识别的目的。霍夫变换是从光学图像中识别特定形状的目标的基本方法之一，而扩展后用于目标特征提取的霍夫变换主要是将图像从空间域变换到由不同的基本形状组成的特征空间中，利用不同几何形状的系数构成目标的特征。
(2)不变矩特征(Moment Feature)：不变矩是光学图像处理中目标检测、识别的一种重要方法。图像的中心矩和原点矩可以区分目标在成像平面投影的几何形状信息，但投影面的几何形状不具有尺度、旋转或者仿射不变性。
(3)全局投影特征：将图像分别投影到几个参考方向，并且只取与参考方向垂直的笔划向参考方向投影。该方法与基于结构特征的笔划提取方法相比，运算简单，能够实现快速提取。全局投影特征能够在一定程度在上反映整个汉字复杂程度，笔划的主要方向以及笔划之间可能存在的连接关系等。为简化计算，通常取四个方向的投影，即横方向、竖方向、正45度方向以及负45度方向。
(4)背景特征：背景部分和汉字的笔划也可以作为汉字字符图像的一种全局特征。通常选取位于图像两对角线上的空白点(非笔划点)统计这些点在字符的各个方向的笔划密度作为图像的全局背景特征。
汉字字符的局部特征需要先将图像划分为不同的局部区域，在不同区域范围内统计 图像的特征。图像的特征可以是前述的全局特征中的一种或者几种的组合。局部特征提取的关键点在与局部区域的划分方法。局部区域的划分方法主要有网格法(包括固定网格法和弹性网格法)、细胞划分方法、方向线素划分方法、四角特征划分方法以及Gabor划分方法等。
-字符分类器
经过图像预处理得到的单个字符图像经过归一化处理、字符特征提取后的字符特征向量需要送入到汉字分类器中才能完成快递信息的提取工作。分类器即将输入的未知样本区分为不同的种类，完成待识别样本与样本所属类别的一一对应的任务。图像分类器主要包括欧氏距离分类器、神经网络分类器、支持向量机分类器以及遗传算法分类器几类。
欧氏距离分类器，距离分类器是最简单和直观的分类方法，以高维空间中点的距离作为样本相似度量的主要依据。距离值越小，表示待测样本之间的相似程度越高。
支持向量机分类器，神经网络具有较好的曲线拟合能力和目标的分类能力，在目标的识别检测中具有大量的应用。神经网络的缺点也非常明显。神经网络目前的结构没有可靠的规则，因此算法的收敛速度非常慢，网络的初值选择对算法的性能影响较大，算法容易收敛到极小值等缺点。支持向量机使用了一种特殊类型的操平面。这种超平面被称为最优分类超平面。
-S600：将所述快递信息录入所述寄件页面的相应位置；
-S700：存储所述快递信息。
参考图4，本发明还公开了一种基于智能终端的快递信息录入系统100，所述快递信息录入系统100包括：页面进入模块11、图像获取模块12、图像预处理模块13、图像分割模块14、信息识别模块15、信息录入模块16；
所述页面进入模块11，进入所述智能终端的寄件页面；
所述图像获取模块12，与所述页面进入模块11通信连接，获取一包含快递信息的信息图像；
所述图像预处理模块13，与所述图像获取模块12通信连接，对所述信息图像进行预处理；
所述图像分割模块14，与所述图像预处理模块13通信连接，对预处理后的所述信息图像进行版面分析及区域分割；
所述信息识别模块15，与所述图像分割模块14通信连接，对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息；
所述信息录入模块16，与所述页面进入模块11、信息识别模块15通信连接，将所述快递信息录入所述寄件页面的相应位置。
在一优选实施例中，所述图像获取模块12，开启所述智能终端的摄像头，拍摄并获取一包含快递信息的信息图像；
和/或
所述图像获取模块12，调用所述智能终端的相册应用，获取一包含快递信息的信息图像。
在一优选实施例中，所述图像预处理模块13包括：图像预处理单元、倾斜度校正单元；
所述图像预处理单元，将所述信息图像进行二值化处理或灰度级别处理；
所述倾斜度校正单元，以所述信息图像的边沿或文字的行向为基准进行倾斜度校正。
在一优选实施例中，所述信息识别模块15包括：字符分离单元、特征提取单元、字符匹配单元；
所述字符分离单元，将所述信息图像切割为图像行，从所述图像行中分离出单个字符；
所述特征提取单元，与所述字符分离单元通信连接，从所述单个字符上提取统计特征或结构特征，包括细化和归一化；
所述字符匹配单元，与所述特征提取单元通信连接，从学习得到的特征库中找到与所述单个字符相似度最高的字符类。
在一优选实施例中，所述快递信息录入系统100还包括信息存储模块17，与所述信息识别模块16通信连接，存储所述快递信息。
应当注意的是，本发明的实施例有较佳的实施性，且并非对本发明作任何形式的限制，任何熟悉该领域的技术人员可能利用上述揭示的技术内容变更或修饰为等同的有效实施例，但凡未脱离本发明技术方案的内容，依据本发明的技术实质对以上实施例所作的任何修改或等同变化及修饰，均仍属于本发明技术方案的范围内。

Claims (10)

  1. 一种基于智能终端的快递信息录入方法，其特征在于，包括以下步骤：
    进入所述智能终端的寄件页面；
    获取一包含快递信息的信息图像；
    对所述信息图像进行预处理；
    对预处理后的所述信息图像进行版面分析及区域分割；
    对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息；
    将所述快递信息录入所述寄件页面的相应位置。
  2. 如权利要求1所述的快递信息录入方法，其特征在于，
    获取一包含快递信息的信息图像的步骤包括：
    开启所述智能终端的摄像头，拍摄并获取一包含快递信息的信息图像；
    和/或
    调用所述智能终端的相册应用，获取一包含快递信息的信息图像。
  3. 如权利要求1所述的快递信息录入方法，其特征在于，
    对所述信息图像进行预处理的步骤包括：
    将所述信息图像进行二值化处理或灰度级别处理；
    以所述信息图像的边沿或文字的行向为基准进行倾斜度校正。
  4. 如权利要求1所述的快递信息录入方法，其特征在于，
    对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息的步骤包括：
    将所述信息图像切割为图像行，从所述图像行中分离出单个字符；
    从所述单个字符上提取统计特征或结构特征，包括细化和归一化；
    从学习得到的特征库中找到与所述单个字符相似度最高的字符类。
  5. 如权利要求1所述的快递信息录入方法，其特征在于，
    所述快递信息录入方法还包括：
    存储所述快递信息。
  6. 一种基于智能终端的快递信息录入系统，其特征在于，
    所述快递信息录入系统包括：页面进入模块、图像获取模块、图像预处理模块、图像分割模块、信息识别模块、信息录入模块；
    所述页面进入模块，进入所述智能终端的寄件页面；
    所述图像获取模块，与所述页面进入模块通信连接，获取一包含快递信息的信息图像；
    所述图像预处理模块，与所述图像获取模块通信连接，对所述信息图像进行预处理；
    所述图像分割模块，与所述图像预处理模块通信连接，对预处理后的所述信息图像进行版面分析及区域分割；
    所述信息识别模块，与所述图像分割模块通信连接，对区域分割后的所述信息图像的各个区域进行文字识别，提取所述快递信息；
    所述信息录入模块，与所述页面进入模块、信息识别模块通信连接，将所述快递信息录入所述寄件页面的相应位置。
  7. 如权利要求6所述的快递信息录入系统，其特征在于，
    所述图像获取模块，开启所述智能终端的摄像头，拍摄并获取一包含快递信息的信息图像；
    和/或
    所述图像获取模块，调用所述智能终端的相册应用，获取一包含快递信息的信息图像。
  8. 如权利要求6所述的快递信息录入系统，其特征在于，
    所述图像预处理模块包括：图像预处理单元、倾斜度校正单元；
    所述图像预处理单元，将所述信息图像进行二值化处理或灰度级别处理；
    所述倾斜度校正单元，以所述信息图像的边沿或文字的行向为基准进行倾斜度校正。
  9. 如权利要求6所述的快递信息录入系统，其特征在于，
    所述信息识别模块包括：字符分离单元、特征提取单元、字符匹配单元；
    所述字符分离单元，将所述信息图像切割为图像行，从所述图像行中分离出单个字符；
    所述特征提取单元，与所述字符分离单元通信连接，从所述单个字符上提取统计特征或结构特征，包括细化和归一化；
    所述字符匹配单元，与所述特征提取单元通信连接，从学习得到的特征库中找到与所述单个字符相似度最高的字符类。
  10. 如权利要求6所述的快递信息录入系统，其特征在于，
    所述快递信息录入系统还包括信息存储模块，与所述信息识别模块通信连接，存储所述快递信息。
PCT/CN2017/105743 2017-10-11 2017-10-11 一种基于智能终端的快递信息录入方法及录入系统 WO2019071476A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2017/105743 WO2019071476A1 (zh) 2017-10-11 2017-10-11 一种基于智能终端的快递信息录入方法及录入系统
CN201780095880.7A CN111213157A (zh) 2017-10-11 2017-10-11 一种基于智能终端的快递信息录入方法及录入系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/105743 WO2019071476A1 (zh) 2017-10-11 2017-10-11 一种基于智能终端的快递信息录入方法及录入系统

Publications (1)

Publication Number Publication Date
WO2019071476A1 true WO2019071476A1 (zh) 2019-04-18

Family

ID=66100310

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/105743 WO2019071476A1 (zh) 2017-10-11 2017-10-11 一种基于智能终端的快递信息录入方法及录入系统

Country Status (2)

Country Link
CN (1) CN111213157A (zh)
WO (1) WO2019071476A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443538A (zh) * 2019-06-20 2019-11-12 苏州视印智能系统有限公司 一种收件自动信息识别系统及方法
CN114780685A (zh) * 2022-04-28 2022-07-22 贵州电网有限责任公司 一种自动识别缺陷信息录入情况并通过无人机补录的方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5465304A (en) * 1992-04-06 1995-11-07 Ricoh Corporation Segmentation of text, picture and lines of a document image
CN104408450A (zh) * 2014-11-21 2015-03-11 深圳天源迪科信息技术股份有限公司 身份证识别方法、装置及系统
CN106203416A (zh) * 2016-07-08 2016-12-07 武汉工程大学 一种基于扫描图像的快递单据信息录入方法及录入装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100655916B1 (ko) * 2004-07-19 2006-12-08 한국과학기술원 방대한 데이터의 디지털화를 위한 문서영상처리 및검증시스템 및 그 방법
CN100362525C (zh) * 2005-06-06 2008-01-16 英华达(上海)电子有限公司 手机中使用图像识别进行名片信息的采集与录入的方法
CN104881770A (zh) * 2015-06-03 2015-09-02 秦志勇 一种快递单信息识别系统和方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5465304A (en) * 1992-04-06 1995-11-07 Ricoh Corporation Segmentation of text, picture and lines of a document image
CN104408450A (zh) * 2014-11-21 2015-03-11 深圳天源迪科信息技术股份有限公司 身份证识别方法、装置及系统
CN106203416A (zh) * 2016-07-08 2016-12-07 武汉工程大学 一种基于扫描图像的快递单据信息录入方法及录入装置

Also Published As

Publication number Publication date
CN111213157A (zh) 2020-05-29

Similar Documents

Publication Publication Date Title
AU2020200251B2 (en) Label and field identification without optical character recognition (OCR)
CN110569832B (zh) 基于深度学习注意力机制的文本实时定位识别方法
Luo et al. Design and implementation of a card reader based on build-in camera
US20220292862A1 (en) Document spatial layout feature extraction to simplify template classification
US8744196B2 (en) Automatic recognition of images
US7630551B2 (en) Method and system for line extraction in digital ink
Aradhye A generic method for determining up/down orientation of text in roman and non-roman scripts
US20170109610A1 (en) Building classification and extraction models based on electronic forms
EP2881894A2 (en) Scaling mobile check photos to physical dimensions
WO2016131083A1 (en) Identity verification. method and system for online users
US8792730B2 (en) Classification and standardization of field images associated with a field in a form
US20140301608A1 (en) Chemical structure recognition tool
US11574492B2 (en) Efficient location and identification of documents in images
Shafii Optical character recognition of printed persian/arabic documents
CN113011426A (zh) 一种识别证件的方法和装置
CN108090728B (zh) 一种基于智能终端的快递信息录入方法及录入系统
CN114581928A (zh) 一种表格识别方法及系统
WO2019071476A1 (zh) 一种基于智能终端的快递信息录入方法及录入系统
US20230132261A1 (en) Unified framework for analysis and recognition of identity documents
Kumar et al. Online handwritten character recognition for Telugu language using support vector machines
CN114386413A (zh) 处理数字化的手写
CN113887484B (zh) 一种卡片式文件图像识别方法和装置
TWI773444B (zh) 影像識別系統與方法
Tamirat Customers Identity Card Data Detection and Recognition Using Image Processing
CN113538291A (zh) 卡证图像倾斜校正方法、装置、计算机设备和存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17928346

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17928346

Country of ref document: EP

Kind code of ref document: A1