WO2017016240A1 - 一种钞票冠字号识别方法 - Google Patents

一种钞票冠字号识别方法 Download PDF

Info

Publication number
WO2017016240A1
WO2017016240A1 PCT/CN2016/078869 CN2016078869W WO2017016240A1 WO 2017016240 A1 WO2017016240 A1 WO 2017016240A1 CN 2016078869 W CN2016078869 W CN 2016078869W WO 2017016240 A1 WO2017016240 A1 WO 2017016240A1
Authority
WO
WIPO (PCT)
Prior art keywords
character
image
recognition result
crown
banknote
Prior art date
Application number
PCT/CN2016/078869
Other languages
English (en)
French (fr)
Inventor
王丹丹
肖助明
岳许要
刘道余
Original Assignee
广州广电运通金融电子股份有限公司
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 广州广电运通金融电子股份有限公司 filed Critical 广州广电运通金融电子股份有限公司
Publication of WO2017016240A1 publication Critical patent/WO2017016240A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G07CHECKING-DEVICES
    • G07DHANDLING OF COINS OR VALUABLE PAPERS, e.g. TESTING, SORTING BY DENOMINATIONS, COUNTING, DISPENSING, CHANGING OR DEPOSITING
    • G07D7/00Testing specially adapted to determine the identity or genuineness of valuable papers or for segregating those which are unacceptable, e.g. banknotes that are alien to a currency
    • G07D7/20Testing patterns thereon
    • G07D7/2016Testing patterns thereon using feature extraction, e.g. segmentation, edge detection or Hough-transformation

Definitions

  • the invention relates to the technical field of banknote recognition, and in particular to a method for identifying a banknote crown number.
  • the crown size of the banknote is generally composed of the numbers 0-9 and the letters A to Z.
  • the first digit is a letter
  • the second, third, and fourth digits are numbers or letters
  • the fifth to tenth digits are numbers.
  • the embodiment of the invention provides a method for recognizing a banknote crown number, which can solve the problem that the existing character recognition processing cannot well recognize similar characters.
  • the secondary recognition result is output.
  • the method before the sliding matching in the specific area by using the stroke template, the method further includes:
  • the character image is binarized.
  • the stroke template includes: a stroke horizontal, a stroke vertical, and a stroke point.
  • the similar character group corresponding to the preliminary recognition result includes a first similar character and a second similar character
  • the recognition result of the character image is a first similar character, and if not, the recognition result of the character image is a second similar character.
  • the method before performing the character cutting process on the crown image, the method further includes:
  • the image preprocessing of the image of the crown number specifically includes:
  • extracting the feature vector of the character image specifically includes:
  • the first feature subvector and the second feature subvector are synthesized into the feature vector.
  • the classifier model is composed of an SMO binary classifier, which includes a digital model, an alphabet model, and a digital and alpha model;
  • the classifier model has a corresponding relationship with the position of the character image on the banknote crown number.
  • the banknote is RMB
  • the similar character group includes a first similar character group 0 and D, a second similar character group 8 and B, and a Three similar character groups 1 and I, fourth similar character group 2 and Z, or fifth similar character group O and D;
  • the first digit of the banknote crown number corresponds to the letter model
  • the fifth digit to the tenth digit correspond to the digital model
  • the second digit to the fourth digit correspond to the number and letter model.
  • the determining whether the preliminary recognition result falls into a preset similar character group is specifically:
  • scaling all of the character images to a preset same size is specifically:
  • the secondary recognition of the character image specifically includes the following steps: according to the preliminary recognition result Obtaining a preset specific area of the character image; acquiring a preset stroke template of the character image according to the preliminary recognition result; performing sliding matching in the specific area by the stroke template, and matching the successfully matched character
  • the maximum number of pixels of the image is taken as the maximum matching value; according to the maximum matching value and the preset threshold value Said character image similar to the character recognition result in the group,
  • the character image is initially identified according to the feature vector and the pre-trained classifier model, and then the character image falling into the similar character group is secondarily recognized, and the whole is combined.
  • the analysis of the statistical features of the bureau and the characteristics of the local structural feature analysis can identify similar characters in the crown size of the banknote, greatly improving the accuracy of the crown identification of the banknote.
  • FIG. 1 is a flow chart of an embodiment of a banknote crown number identification method according to an embodiment of the present invention
  • FIG. 2 is a flow chart of another embodiment of a banknote crown number identification method according to an embodiment of the present invention.
  • FIG. 3 is a schematic diagram showing the identification of a banknote crown number recognition method according to an embodiment of the present invention.
  • FIG. 4 is a schematic exploded perspective view of an embodiment of the present invention.
  • Figure 5 is a schematic diagram of a stroke template
  • FIG. 6 is a schematic diagram of a template allocation area of similar characters in an embodiment of the present invention.
  • the embodiment of the invention provides a method for recognizing a banknote crown number, which is used for solving the problem that the existing character recognition processing cannot identify similar characters well.
  • an embodiment of a banknote crown number identification method includes:
  • the crown image After acquiring the crown image of the banknote, the crown image may be subjected to character cutting processing to obtain a plurality of character images.
  • all of the character images can be scaled to a preset same size.
  • the feature vector of the character image can be extracted.
  • the character image After extracting the feature vector of the character image, the character image may be subjected to character recognition according to the feature vector and the pre-trained classifier model to obtain a preliminary recognition result.
  • step 106 determine whether the preliminary recognition result falls into a preset similar character group, and if so, step 108 is performed; if not, step 107 is performed;
  • step 108 is performed, and if not, step 107 is performed.
  • the preliminary recognition result does not fall into the preset similar character group. If the preliminary recognition result does not fall into the preset similar character group, the preliminary recognition result is output.
  • the character image needs to be secondarily recognized.
  • the preset specific region of the character image may be acquired according to the preliminary recognition result.
  • the preset stroke template of the character image can be obtained according to the preliminary recognition result.
  • the stroke matching may be performed in the specific area by the stroke template, and the maximum number of pixels of the character image that is successfully matched is used as the maximum matching value.
  • Step 111 Obtain a recognition result of the character image in the similar character group according to the maximum matching value and a preset threshold value, as a secondary recognition result;
  • the recognition result of the character image in the similar character group may be obtained according to the maximum matching value and the preset threshold value as a secondary recognition result.
  • the secondary recognition of the character image specifically includes the following steps: acquiring a preset specific region of the character image according to the preliminary recognition result; Initially obtaining a preset stroke template of the character image; performing sliding matching in the specific area by using the stroke template, and using the maximum number of pixels of the character image that is successfully matched as a maximum matching value; according to the maximum matching value and the pre- Setting a threshold value to obtain a recognition result of the character image in the similar character group Second recognition result
  • the model model performs preliminary recognition on the character image, and then performs secondary recognition on the character image falling into the similar character group.
  • the similar character in the banknote crown number can be identified. Greatly improve the accuracy of the banknote crown identification.
  • FIG. 2 another embodiment of the banknote crown number identification method in the embodiment of the present invention includes:
  • the image of the crown of the banknote can be obtained by image scanning or the like.
  • the image of the crown number in this embodiment may be a grayscale image formed by a white light source, or may be a grayscale image of a single channel of a color image.
  • image preprocessing can be performed on the crown image.
  • the image preprocessing of the image of the crown number includes: performing denoising processing on the image of the crown size, or performing image rectification processing on the image of the crown size. It can be understood that the pre-processed crown image can be better cut out characters, avoiding the problem that the cutting characters are offset or error and the recognition result is inaccurate.
  • character cutting processing may be performed on the crown image to obtain a plurality of character images. It should be noted that the binarized character image can be horizontally and vertically projected, and then each character is cut out.
  • all of the character images may be scaled to a preset same size, for example, all of the character images are scaled to a rectangular image of the same size W*H according to interpolation, where W is the width of the character image , H is the height of the character image.
  • the feature vector of the character image can be extracted.
  • the extracting the feature vector of the character image may specifically include:
  • the character image After extracting the feature vector of the character image, the character image may be subjected to character recognition according to the feature vector and the pre-trained classifier model to obtain a preliminary recognition result.
  • the classifier model is composed of an SMO two classifier, which includes a digital model and a letter model, and the classifier model has a corresponding relationship with the position of the character image on the banknote crown number.
  • the similar character group includes a first similar character group 0 and D, a second similar character group 8 and B, a third similar character group 1 and I, and a fourth similar character group 2 and Z, Or a fifth similar character group O and D; the first digit of the banknote crown number corresponds to the letter model, the fifth to tenth digits correspond to the digital model, the second digit to the fourth digit and the number and letter model correspond.
  • step 207 it is determined whether the preliminary recognition result falls into a preset similar character group, and if so, step 209 is performed, and if not, step 208 is performed;
  • step 207 is equivalent to determining whether the preliminary recognition result is 0, D, 8, B, 1, I, 2, Z or O, and if yes, executing step 209, if no Then, step 208 is performed.
  • the preliminary recognition result may be considered to be accurate, there is no approximation, and the preliminary recognition result is output as the final recognition result.
  • the preliminary recognition result falls within a preset similar character group, the preliminary recognition result may be considered to be suspicious, there may be an approximation, and secondary recognition is required.
  • the character image is binarized.
  • the preset specific region of the character image may be acquired according to the preliminary recognition result. It can be understood that the preset specific area is determined for the preliminary recognition result, for example, if the preliminary recognition result is the number 1, the specific area is the upper position of "1".
  • the preset stroke region of the character image may be acquired according to the preliminary recognition result, while the preset specific region of the character image is acquired according to the preliminary recognition result. It can be understood that the stroke template is determined according to the preliminary recognition result, for example, if the preliminary recognition result is D, due to 0 and D It is a similar character, and D has a vertical and 0 is not, so at this time the stroke template selects the stroke vertical.
  • the stroke template may include: a stroke stroke, a stroke vertical, and a stroke point.
  • the stroke matching may be performed in the specific region by using the stroke template, and the maximum number of pixels of the character image that is successfully matched is used as the maximum matching value.
  • the recognition result of the character image in the similar character group may be obtained according to the maximum matching value and the preset threshold value as a secondary recognition result.
  • the similar character group corresponding to the preliminary recognition result includes the first similar character and the second similar character
  • the character image is obtained in the similar character group according to the maximum matching value and the preset threshold.
  • the recognition result of the character image is a first similar character, and if not, the recognition result of the character image is a second similar character.
  • the secondary recognition result can be considered to be accurate and reliable, and the secondary recognition result is output as the final recognition result.
  • Step 1 Read the grayscale image of the crown size area
  • Step 2 Pre-processing such as image denoising and image rotation correction on the crown size area
  • Step 3 Horizontally and vertically project the binarized character image, cut out each character, and use the interpolation method to scale the image to a uniform size W*H.
  • Step 4 Extract the feature vector
  • Step 4-1 Calculate the character gradient map to obtain a gradient matrix
  • Step 4-2 Feature matrix extraction, the gradient matrix G is decomposed in the standard eight directions, as shown in FIG.
  • pi is the decomposition value of (gx, gy) in direction i.
  • Step 4-3 Feature matrix processing, setting the weight matrix w i,j represents the weight of the feature. This embodiment selects That is, the average filtering is performed on each of the eight directions of P.
  • n d, (i, j) is the number of gradients of the pixel (i, j) decomposed into the d direction
  • D_max_project is the direction in which N_max_project takes the maximum value
  • p d, (i, j) is the decomposition value of the gradient of the pixel (i, j) in the d direction.
  • D_max_grad is the direction in which V_max_grad takes the maximum value.
  • D_max_project, N_max_projec, D_max_grad, V_max_grad generate a feature sub-vector F2.
  • F1 and F2 synthesize feature vectors F.
  • Step 5 Identification of non-similar characters and similar characters
  • the classifier model used in this step consists of a pre-trained SMO binary classifier.
  • the features of all digital and alpha training samples are extracted as described in step 4.
  • Each character is tagged, and the number of the tags is 0 to 9 according to the numerical value, and the letters (except V) are sequentially sorted according to the order of 10 to 35.
  • Two different characters train one SMO two classifier, and the second classifier model includes positive and negative sample labels: pos and neg, threshold b and weight array A. All two classifiers with pos and neg less than 10 form a digital model; all two classifiers with pos and neg greater than or equal to 10 form a letter model; all two classifiers form a digital and alphabetic model.
  • the product of the feature vector F of the character to be recognized and the weight class array A of the two classifiers is calculated, and the value S, S summed with the threshold b is greater than 0 to return the positive label pos, and otherwise, the negative label neg is returned.
  • Each of the two classifiers casts 1 vote for the character class represented by the returned tag, and all the two classifiers constituting the classifier model vote for the recognized characters in turn, and finally counts the number of votes for each class, and the class with the most votes is the character to be recognized. Category.
  • the first digit is a letter
  • the last six digits are numbers
  • the second, third, and fourth digits are numbers or letters.
  • a digital model, an alphabet model or a numerical and alphabetic model is selected according to the position of the character to be recognized in the crown number sequence. The first digit uses the alphabet model, the last six digits use the numeric model, and the second, third, and fourth digits use the numeric and alphabetic model.
  • the SMO recognition result is the category with the final recognized character. If the recognition result is a character in a similar character group, it needs to be recognized again by using the stroke template matching.
  • the template corresponding to the set of characters is used to perform template matching in the matching area, and the matching result is the final recognition result.
  • the method involved in this step includes:
  • the character image is binarized using the Otsu algorithm.
  • a stroke template is a rectangular area that is similar in size to the stroke of a character and is used to distinguish between strokes with similar characters.
  • D has a stroke vertical and O is not, so a rectangular area similar to the vertical dimension of the stroke of D can be set as a vertical template of the stroke.
  • Template 1 represents the stroke of the stroke
  • template 2 represents the stroke vertical
  • template 3 represents the stroke point.
  • the template matching area of each group of similar characters is set according to the position and size of the area where the similar characters are different.
  • O and D, 8 and B are different in the left half
  • 2 and Z are different in the upper half
  • 1 and I are different in the upper half.
  • the separable domains of O and D, 8 and B are left half images.
  • the separable domains of 2 and Z are the upper 1/3 images
  • the separable domains of 1 and I are the upper intermediate positions.
  • the preset stroke template and the matching area corresponding to the similar characters are selected according to the initial recognition result of the characters described above. Assuming that the recognition result is 2 and Z, the stroke template 1 is selected, and the matching area is the upper half of the character.
  • the stroke template slides from the upper left corner to the lower right corner of the matching area corresponding to the character, calculates the matching result of each sliding, and counts the maximum matching value max_I of the template m in the matching area, as shown in equation (5). .
  • x_Start, x_End, y_Start, y_End are four vertices that can be divided into domains.
  • Max_I table Shows the largest match value.
  • the disclosed system, apparatus, and method may be implemented in other manners.
  • the device embodiments described above are merely illustrative.
  • the division of the unit is only a logical function division.
  • there may be another division manner for example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be in an electrical, mechanical or other form.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. in.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium.
  • the technical solution of the present invention which is essential or contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product stored in a storage medium.
  • a number of instructions are included to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention.
  • the foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and the like. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Character Discrimination (AREA)

Abstract

一种钞票冠字号识别方法,用于解决现有字符识别处理无法很好地识别相似字符的问题。包括:获取钞票的冠字号图像(101);对所述冠字号图像进行字符切割处理,得到多个字符图像(102);将所有所述字符图像缩放成预设的同一尺寸(103);提取所述字符图像的特征向量(104);根据所述特征向量和预先训练的分类器模型对所述字符图像进行字符识别,得到初步识别结果(105);判断所述初步识别结果是否落入预设的相似字符组中(106),若是,则对所述字符图像进行二次识别(108),若否,则输出所述初步识别结果(107)。

Description

一种钞票冠字号识别方法
本申请要求于2015年7月24日提交中国专利局、申请号为201510443057.3、发明名称为“一种钞票冠字号识别方法”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本发明涉及钞票识别技术领域,尤其涉及一种钞票冠字号识别方法。
背景技术
随着光学字符识别技术在人类生产、生活中广泛应用,钞票冠字号识别已经成为金融领域一种防止***的重要手段,光学字符自动识别技术越来越受到研究者的关注。
钞票冠字号一般由数字0~9以及字母A~Z组合而成。以目前人民币冠字号序列规则来说,第1位为字母,第2、3、4位为数字或字母,第5到10位为数字。
在字符识别处理中,字符特征提取是非常重要的步骤之一,特别是在钞票识别领域中相似字符(O和0,1和I等)的特征相似度极高。然而基于传统的局部结构特征或全局的统计特征,都不能很好的解决相似字符的问题,因此本领域技术人员亟需寻找一种可识别相似字符的钞票冠字号识别方法。
发明内容
本发明实施例提供了一种钞票冠字号识别方法,能够解决现有字符识别处理无法很好地识别相似字符的问题。
本发明实施例提供的一种钞票冠字号识别方法,包括:
获取钞票的冠字号图像;
对所述冠字号图像进行字符切割处理,得到多个字符图像;
将所有所述字符图像缩放成预设的同一尺寸;
提取所述字符图像的特征向量;
根据所述特征向量和预先训练的分类器模型对所述字符图像进行字符识别,得到初步识别结果;
判断所述初步识别结果是否落入预设的相似字符组中,若是,则对所述字符图像进行二次识别,若否,则输出所述初步识别结果;
所述对所述字符图像进行二次识别具体包括以下步骤:
根据所述初步识别结果获取所述字符图像的预设特定区域;
根据所述初步识别结果获取所述字符图像的预设笔画模板;
通过所述笔画模板在所述特定区域内进行滑动匹配,将匹配成功的所述字符图像的像素数最大值作为最大匹配值;
根据所述最大匹配值和预设的阈值得到所述字符图像在所述相似字符组中的识别结果,作为二次识别结果;
输出所述二次识别结果。
可选地,通过所述笔画模板在所述特定区域内进行滑动匹配之前还包括:
将所述字符图像进行二值化处理。
可选地,所述笔画模板包括:笔画横、笔画竖以及笔画点。
可选地,与所述初步识别结果对应的所述相似字符组包括第一相似字符和第二相似字符;
所述根据所述最大匹配值和预设的阈值得到所述字符图像在所述相似字符组中的识别结果具体为:
判断所述最大匹配值是否大于或等于预设的阈值,若是,则所述字符图像的识别结果为第一相似字符,若否,则所述字符图像的识别结果为第二相似字符。
可选地,对所述冠字号图像进行字符切割处理之前还包括:
对所说冠字号图像进行图像预处理;
所述对所说冠字号图像进行图像预处理具体包括:
对所述冠字号图像进行去噪处理;
或,对所述冠字号图像进行图像纠偏处理。
可选地,提取所述字符图像的特征向量具体包括:
计算所述字符图像的梯度图,得到梯度矩阵;
在标准八方向分解所述梯度矩阵,得到特征矩阵;
将所述特征矩阵与预设的权重矩阵进行卷积生成第一特征子向量;
根据所述梯度矩阵在标准八方向上分解次数最多的方向及其最大数量,和分解值累加最大的方向及最大累加值,生成第二特征子向量;
将所述第一特征子向量和所述第二特征子向量合成为所述特征向量。
可选地,所述分类器模型由SMO二分类器组成,其包括数字模型、字母模型、以及数字和字母模型;
所述分类器模型与所述字符图像在钞票冠字号上的位置存在对应关系。
可选地,所述钞票为人民币;
所述相似字符组包括第一相似字符组0和D、第二相似字符组8和B、第 三相似字符组1和I、第四相似字符组2和Z、或者第五相似字符组O和D;
所述钞票冠字号的第一位与所述字母模型对应,第五位至第十位与所述数字模型对应,第二位至第四位与所述数字和字母模型对应。
可选地,所述判断所述初步识别结果是否落入预设的相似字符组中具体为:
判断所述初步识别结果是否为0、D、8、B、1、I、2、Z或者O。
可选地,将所有所述字符图像缩放成预设的同一尺寸具体为:
根据插值法将所有所述字符图像缩放成同一尺寸的矩形图像。
从以上技术方案可以看出,本发明实施例具有以下优点:
本发明实施例中,首先,获取钞票的冠字号图像;对所述冠字号图像进行字符切割处理,得到多个字符图像;将所有所述字符图像缩放成预设的同一尺寸;然后,提取所述字符图像的特征向量;根据所述特征向量和预先训练的分类器模型对所述字符图像进行字符识别,得到初步识别结果;最后,判断所述初步识别结果是否落入预设的相似字符组中,若是,则对所述字符图像进行二次识别,若否,则输出所述初步识别结果;其中,所述对所述字符图像进行二次识别具体包括以下步骤:根据所述初步识别结果获取所述字符图像的预设特定区域;根据所述初步识别结果获取所述字符图像的预设笔画模板;通过所述笔画模板在所述特定区域内进行滑动匹配,将匹配成功的所述字符图像的像素数最大值作为最大匹配值;根据所述最大匹配值和预设的阈值得到所述字符图像在所述相似字符组中的识别结果,作为二次识别结果;输出所述二次识别结果。在本发明实施例中,根据所述特征向量和预先训练的分类器模型对字符图像进行初步识别,然后对落入相似字符组的字符图像进行二次识别,结合了全 局统计特征的分析和局部结构特征分析的特点,可以识别钞票冠字号中的相似字符,极大提升钞票冠字号识别的准确性。
附图说明
图1为本发明实施例中一种钞票冠字号识别方法一个实施例流程图;
图2为本发明实施例中一种钞票冠字号识别方法另一个实施例流程图;
图3为本发明实施例中一种钞票冠字号识别方法的识别原理图;
图4为本发明实施例中梯度分解示意图;
图5为笔画模板示意图;
图6为本发明实施例中相似字符的模板分配区域的示意图。
具体实施方式
本发明实施例提供了一种钞票冠字号识别方法,用于解决现有字符识别处理无法很好地识别相似字符的问题。
为使得本发明的发明目的、特征、优点能够更加的明显和易懂,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,下面所描述的实施例仅仅是本发明一部分实施例,而非全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例,都属于本发明保护的范围。
请参阅图1,本发明实施例中一种钞票冠字号识别方法一个实施例包括:
101、获取钞票的冠字号图像;
首先,需要获取钞票的冠字号图像。
102、对该冠字号图像进行字符切割处理,得到多个字符图像;
在获取钞票的冠字号图像之后,可以对该冠字号图像进行字符切割处理,得到多个字符图像。
103、将所有该字符图像缩放成预设的同一尺寸;
在得到多个字符图像之后,可以将所有该字符图像缩放成预设的同一尺寸。
104、提取该字符图像的特征向量;
在将所有该字符图像缩放成预设的同一尺寸之后,可以提取该字符图像的特征向量。
105、根据该特征向量和预先训练的分类器模型对该字符图像进行字符识别,得到初步识别结果;
在提取该字符图像的特征向量之后,可以根据该特征向量和预先训练的分类器模型对该字符图像进行字符识别,得到初步识别结果。
106、判断该初步识别结果是否落入预设的相似字符组中,若是,则执行步骤108,若否,则执行步骤107;
在得到初步识别结果之后,可以判断该初步识别结果是否落入预设的相似字符组中,若是,则执行步骤108,若否,则执行步骤107。
107、输出该初步识别结果;
若该初步识别结果未落入预设的相似字符组中,则输出该初步识别结果。
108、根据该初步识别结果获取该字符图像的预设特定区域;
若该初步识别结果落入预设的相似字符组中,则需要对该字符图像进行二次识别,首先,可以根据该初步识别结果获取该字符图像的预设特定区域。
109、根据该初步识别结果获取该字符图像的预设笔画模板;
并且,可以根据该初步识别结果获取该字符图像的预设笔画模板。
110、通过该笔画模板在该特定区域内进行滑动匹配,将匹配成功的该字符图像的像素数最大值作为最大匹配值;
在得到该特定区域和该笔画模板之后,可以通过该笔画模板在该特定区域内进行滑动匹配,将匹配成功的该字符图像的像素数最大值作为最大匹配值。
111、根据该最大匹配值和预设的阈值得到该字符图像在该相似字符组中的识别结果,作为二次识别结果;
在得到该最大匹配值之后,可以根据该最大匹配值和预设的阈值得到该字符图像在该相似字符组中的识别结果,作为二次识别结果。
112、输出该二次识别结果。
最后,输出该二次识别结果。
本实施例中,首先,获取钞票的冠字号图像;对该冠字号图像进行字符切割处理,得到多个字符图像;将所有该字符图像缩放成预设的同一尺寸;然后,提取该字符图像的特征向量;根据该特征向量和预先训练的分类器模型对该字符图像进行字符识别,得到初步识别结果;最后,判断该初步识别结果是否落入预设的相似字符组中,若是,则对该字符图像进行二次识别,若否,则输出该初步识别结果;其中,该对该字符图像进行二次识别具体包括以下步骤:根据该初步识别结果获取该字符图像的预设特定区域;根据该初步识别结果获取该字符图像的预设笔画模板;通过该笔画模板在该特定区域内进行滑动匹配,将匹配成功的该字符图像的像素数最大值作为最大匹配值;根据该最大匹配值和预设的阈值得到该字符图像在该相似字符组中的识别结果,作为二次识别结果;输出该二次识别结果。在本实施例中,根据该特征向量和预先训练的分类 器模型对字符图像进行初步识别,然后对落入相似字符组的字符图像进行二次识别,结合了全局统计特征的分析和局部结构特征分析的特点,可以识别钞票冠字号中的相似字符,极大提升钞票冠字号识别的准确性。
为便于理解,下面对本发明实施例中的一种钞票冠字号识别方法进行详细描述,请参阅图2,本发明实施例中一种钞票冠字号识别方法另一个实施例包括:
201、获取钞票的冠字号图像;
首先,可以获取钞票的冠字号图像。可以理解的是,可以通过图像扫描等方式获取需要识别的钞票的冠字号图像。
需要说明的是,本实施例中的冠字号图像可以是白光光源形成的灰度图,也可以是彩色图像的单一通道的灰度图。
202、对所说冠字号图像进行图像预处理;
在获取钞票的冠字号图像之后,可以对所说冠字号图像进行图像预处理。其中,该对所说冠字号图像进行图像预处理具体包括:对该冠字号图像进行去噪处理,或对该冠字号图像进行图像纠偏处理。可以理解的是,经过预处理的冠字号图像可以更好地被切割出字符,避免切割字符出现偏移或者误差而导致识别结果不准确的问题。
203、对该冠字号图像进行字符切割处理,得到多个字符图像;
在对所说冠字号图像进行图像预处理之后,可以对该冠字号图像进行字符切割处理,得到多个字符图像。需要说明的是,可以对二值化后的字符图像进行水平垂直投影,然后切割出每一个字符。
204、将所有该字符图像缩放成预设的同一尺寸;
在得到多个字符图像之后,可以将所有该字符图像缩放成预设的同一尺寸,比如,根据插值法将所有该字符图像缩放成同一尺寸的矩形图像W*H,其中W为字符图像的宽度,H为字符图像的高度。
205、提取该字符图像的特征向量;
在将所有该字符图像缩放成预设的同一尺寸之后,可以提取该字符图像的特征向量。其中,提取该字符图像的特征向量具体可以包括:
A、计算该字符图像的梯度图,得到梯度矩阵;
B、在标准八方向分解该梯度矩阵,得到特征矩阵;
C、将该特征矩阵与预设的权重矩阵进行卷积生成第一特征子向量;
D、根据该梯度矩阵在标准八方向上分解次数最多的方向及其最大数量,和分解值累加最大的方向及最大累加值,生成第二特征子向量;
E、将该第一特征子向量和该第二特征子向量合成为该特征向量。
206、根据该特征向量和预先训练的分类器模型对该字符图像进行字符识别,得到初步识别结果;
在提取该字符图像的特征向量之后,可以根据该特征向量和预先训练的分类器模型对该字符图像进行字符识别,得到初步识别结果。
需要说明的是,该分类器模型由SMO二分类器组成,其包括数字模型和字母模型,该分类器模型与该字符图像在钞票冠字号上的位置存在对应关系。比如,假设该钞票为人民币,则该相似字符组包括第一相似字符组0和D、第二相似字符组8和B、第三相似字符组1和I、第四相似字符组2和Z、或者第五相似字符组O和D;该钞票冠字号的第一位与该字母模型对应,第五位至第十位与该数字模型对应,第二位至第四位与该数字和字母模型对应。
207、判断该初步识别结果是否落入预设的相似字符组中,若是,则执行步骤209,若否,则执行步骤208;
在得到该初步识别结果之后,可以判断该初步识别结果是否落入预设的相似字符组中,若是,则执行步骤209,若否,则执行步骤208。承接上述步骤206的举例,可以知道的是,步骤207相当于判断该初步识别结果是否为0、D、8、B、1、I、2、Z或者O,若是,则执行步骤209,若否,则执行步骤208。
208、输出该初步识别结果;
若该初步识别结果未落入预设的相似字符组中,则可以认为该初步识别结果准确,不存在近似,输出该初步识别结果作为最终的识别结果。
209、将该字符图像进行二值化处理;
若该初步识别结果落入预设的相似字符组中,则可以认为该初步识别结果存疑,可能存在近似,需要进行二次识别。首先,将该字符图像进行二值化处理。
210、根据该初步识别结果获取该字符图像的预设特定区域;
在将该字符图像进行二值化处理之后,可以根据该初步识别结果获取该字符图像的预设特定区域。可以理解的是,该预设特定区域是针对该初步识别结果进行确定的,比如若该初步识别结果为数字1,则该特定区域为“1”的上部位置。
211、根据该初步识别结果获取该字符图像的预设笔画模板;
在根据该初步识别结果获取该字符图像的预设特定区域的同时,还可以根据该初步识别结果获取该字符图像的预设笔画模板。可以理解的是,该笔画模板是根据该初步识别结果来确定的,比如若该初步识别结果为D,由于0与D 为相似字符,而D有一竖而0没有,因此此时该笔画模板选择笔画竖。该笔画模板可以包括:笔画横、笔画竖以及笔画点。
212、通过该笔画模板在该特定区域内进行滑动匹配,将匹配成功的该字符图像的像素数最大值作为最大匹配值;
在获取到该初步识别结果对应的特定区域以及笔画模板之后,可以通过该笔画模板在该特定区域内进行滑动匹配,将匹配成功的该字符图像的像素数最大值作为最大匹配值。
213、根据该最大匹配值和预设的阈值得到该字符图像在该相似字符组中的识别结果,作为二次识别结果;
在得到该最大匹配值之后,可以根据该最大匹配值和预设的阈值得到该字符图像在该相似字符组中的识别结果,作为二次识别结果。
需要说明的是,假设与该初步识别结果对应的该相似字符组包括第一相似字符和第二相似字符,那么该根据该最大匹配值和预设的阈值得到该字符图像在该相似字符组中的识别结果具体为:
判断该最大匹配值是否大于或等于预设的预置,若是,则该字符图像的识别结果为第一相似字符,若否,则该字符图像的识别结果为第二相似字符。
214、输出该二次识别结果。
在得到该二次识别结果之后,由于已经排除了相似字符的情况,因此可以认为该二次识别结果是准确可靠的,输出该二次识别结果作为最终的识别结果。
为便于理解,根据图2所描述的实施例,下面以一个实际应用场景对本发明实施例中的钞票冠字号识别方法进行描述,请参阅图3
步骤1:读入冠字号区域的灰度图;
步骤2:对冠字号区域图像去噪、图像旋转纠偏等预处理;
步骤3:对二值化后的字符图像进行水平垂直投影,切割出每一个字符,使用插值法将图像缩放到统一的尺寸W*H。
步骤4:提取特征向量;
步骤4-1:计算字符梯度图,得梯度矩阵;
使用Sobe算子计算归一化后的字符图像的梯度矩阵,
Figure PCTCN2016078869-appb-000001
步骤4-2:特征矩阵提取,将梯度矩阵G在标准八方向分解,如图4所示。
得到特征矩阵
Figure PCTCN2016078869-appb-000002
其中,pi是(gx,gy)在方向i上的分解值。
步骤4-3:特征矩阵处理,设置权重矩阵
Figure PCTCN2016078869-appb-000003
wi,j表示特征的权重。本实施例选择
Figure PCTCN2016078869-appb-000004
即对P的8个方向分别进行均值滤波。
特征矩阵P与权重W的卷积生成特征子向量F1,
Figure PCTCN2016078869-appb-000005
计算8个方向上分解次数最多的方向D_max_project,及其最大数量N_max_project,和8个方向的分解值累加最大的方向D_max_grad,和最大累加值V__max_grad,计算方法见式(1)—式(4)。
Figure PCTCN2016078869-appb-000006
Figure PCTCN2016078869-appb-000007
Figure PCTCN2016078869-appb-000008
其中,nd,(i,j)是像素(i,j)的梯度分解到d方向的数量,D_max_project是N_max_project取得最大值的方向。
Figure PCTCN2016078869-appb-000009
Figure PCTCN2016078869-appb-000010
其中,pd,(i,j)是像素(i,j)的梯度在d方向的分解值。D_max_grad是V_max_grad取得最大值的方向。
D_max_project、N_max_projec、D_max_grad,V_max_grad生成特征子向量F2。F1和F2合成特征向量F。
步骤5:非相似字符和相似字符组识别
该步骤使用的分类器模型是由预先训练的SMO二分类器组成。根据步骤4所述,提取所有数字和字母训练样本的特征。给每个字符加标签,数字的标签根据数值大小依次为0~9,和字母(V除外)根据排序先后,依次为10~35。 两个不同的字符训练一个SMO二分类器,二分类器模型包括正负样本标签:pos和neg,阈值b和权重数组A。所有pos和neg均小于10的二分类器组成数字模型;所有pos和neg均大于等于10的二分类器组成字母模型;所有的二分类器组成数字和字母模型。
计算待识别字符的特征向量F和二分类器权重数组A的乘积,并与阈值b加和的值S,S大于0返回正标签pos,否则,返回负标签neg。每个二分类器对返回的标签所代表的字符类别投1票,组成分类器模型的所有二分类器依次对待识别字符投票,最后统计每一个类别得票数量,得票最多的类即为待识别字符的类别。
依据人民币冠字号序列规则,第1位为字母,后6位为数字,第2、3、4位为数字或字母。为了提高识别率和分类速度,根据待识别字符在冠字号序列中的位置选择数字模型、字母模型或者数字和字母模型进行分类。第一位使用字母模型,后6位使用数字模型,第2、3、4位使用数字和字母模型。
对于非相似字符,SMO识别结果即为最终带识别字符的类别。若识别结果为相似字符组中的某一个字符,则需要使用笔画模板匹配再次识别。
步骤6:相似字符识别
当使用数字和字母模型识别的结果为某相似字符组中的一个字符时,使用该组字符对应的笔画模板在匹配区域内进行模板匹配,匹配结果为最终识别结果。该步骤涉及的方法包括;
1、二值化
使用Otsu算法对字符图像进行二值化。
2、构造笔画模板
在钞票冠字号中,相似字符包括O(或者0)和D,8和B,1和I,2和Z等。笔画模板是尺寸与字符的笔画相近的矩形区域,用于区分相似字符中有差异的笔画。比如,在O和D的字符图像中,D有一笔画竖,而O没有,因此可以设定与D的笔画竖尺寸相近的矩形区域为笔画竖的模板。依次类推,可构造笔画横、笔画点的模板。如图5所示。模板1表示笔画横,模板2表示笔画竖,模板3表示笔画点。模板m的尺寸Wm*Hm(m=1,2,3)取决于归一化后字符笔画的尺寸。
3、设定模板匹配区域
根据相似字符有差别的区域的位置和大小,设定每组相似字符的模板匹配区域。O和D,8和B左半边不同,2和Z上半边不同,1和I上半边不同。如图6所示,O和D,8和B的可分域为左半边图像。2和Z的可分域为上方1/3的图像,1和I的可分域为上方中间位置。
4、模板匹配
进行笔画模板匹配时,需根据上文所述字符初次识别结果,选择相似字符所对应的预设笔画模板和匹配区域。假设,识别结果为2和Z,则选择笔画模板1,匹配区域为字符上半边。模板匹配时,笔画模板在该字符对应的匹配区域从左上角到右下角滑动,计算每次滑动的匹配结果,并统计模板m在匹配区域内的最大匹配值max_I,如式(5)所示.
Figure PCTCN2016078869-appb-000011
其中,x_Start、x_End,y_Start,y_End为可分域的四个顶点。Max_I表 示最大的匹配值。
Figure PCTCN2016078869-appb-000012
比较max_I与经验阈值T(m)的大小,判断匹配区域内是否存在模板所表示的笔画,如式(6)。例如,对于字符2和Z,匹配区域内存在笔画横,则认为待识别字符为Z,否则为2。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的***,装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的***,装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个***,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本发明各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元 中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(ROM,Read-Only Memory)、随机存取存储器(RAM,Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,以上实施例仅用以说明本发明的技术方案,而非对其限制;尽管参照前述实施例对本发明进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (10)

  1. 一种钞票冠字号识别方法,其特征在于,包括:
    获取钞票的冠字号图像;
    对所述冠字号图像进行字符切割处理,得到多个字符图像;
    将所有所述字符图像缩放成预设的同一尺寸;
    提取所述字符图像的特征向量;
    根据所述特征向量和预先训练的分类器模型对所述字符图像进行字符识别,得到初步识别结果;
    判断所述初步识别结果是否落入预设的相似字符组中,若是,则对所述字符图像进行二次识别,若否,则输出所述初步识别结果;
    所述对所述字符图像进行二次识别具体包括以下步骤:
    根据所述初步识别结果获取所述字符图像的预设特定区域;
    根据所述初步识别结果获取所述字符图像的预设笔画模板;
    通过所述笔画模板在所述特定区域内进行滑动匹配,将匹配成功的所述字符图像的像素数最大值作为最大匹配值;
    根据所述最大匹配值和预设的阈值得到所述字符图像在所述相似字符组中的识别结果,作为二次识别结果;
    输出所述二次识别结果。
  2. 根据权利要求1所述的钞票冠字号识别方法,其特征在于,通过所述笔画模板在所述特定区域内进行滑动匹配之前还包括:
    将所述字符图像进行二值化处理。
  3. 根据权利要求1所述的钞票冠字号识别方法,其特征在于,所述笔画模板包括:笔画横、笔画竖以及笔画点。
  4. 根据权利要求1所述的钞票冠字号识别方法,其特征在于,与所述初步识别结果对应的所述相似字符组包括第一相似字符和第二相似字符;
    所述根据所述最大匹配值和预设的阈值得到所述字符图像在所述相似字符组中的识别结果具体为:
    判断所述最大匹配值是否大于或等于预设的阈值,若是,则所述字符图像的识别结果为第一相似字符,若否,则所述字符图像的识别结果为第二相似字 符。
  5. 根据权利要求1所述的钞票冠字号识别方法,其特征在于,对所述冠字号图像进行字符切割处理之前还包括:
    对所说冠字号图像进行图像预处理;
    所述对所说冠字号图像进行图像预处理具体包括:
    对所述冠字号图像进行去噪处理;
    或,对所述冠字号图像进行图像纠偏处理。
  6. 根据权利要求1所述的钞票冠字号识别方法,其特征在于,提取所述字符图像的特征向量具体包括:
    计算所述字符图像的梯度图,得到梯度矩阵;
    在标准八方向分解所述梯度矩阵,得到特征矩阵;
    将所述特征矩阵与预设的权重矩阵进行卷积生成第一特征子向量;
    根据所述梯度矩阵在标准八方向上分解次数最多的方向及其最大数量,和分解值累加最大的方向及最大累加值,生成第二特征子向量;
    将所述第一特征子向量和所述第二特征子向量合成为所述特征向量。
  7. 根据权利要求1所述的钞票冠字号识别方法,其特征在于,所述分类器模型由SMO二分类器组成,其包括数字模型、字母模型、以及数字和字母模型;
    所述分类器模型与所述字符图像在钞票冠字号上的位置存在对应关系。
  8. 根据权利要求7所述的钞票冠字号识别方法,其特征在于,所述钞票为人民币;
    所述相似字符组包括第一相似字符组0和D、第二相似字符组8和B、第三相似字符组1和I、第四相似字符组2和Z、或者第五相似字符组O和D;
    所述钞票冠字号的第一位与所述字母模型对应,第五位至第十位与所述数字模型对应,第二位至第四位与所述数字和字母模型对应。
  9. 根据权利要求8所述的钞票冠字号识别方法,其特征在于,所述判断所述初步识别结果是否落入预设的相似字符组中具体为:
    判断所述初步识别结果是否为0、D、8、B、1、I、2、Z或者O。
  10. 根据权利要求1至9中任一项所述的钞票冠字号识别方法,其特征在 于,将所有所述字符图像缩放成预设的同一尺寸具体为:
    根据插值法将所有所述字符图像缩放成同一尺寸的矩形图像。
PCT/CN2016/078869 2015-07-24 2016-04-08 一种钞票冠字号识别方法 WO2017016240A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510443057.3A CN105261109B (zh) 2015-07-24 2015-07-24 一种钞票冠字号识别方法
CN201510443057.3 2015-07-24

Publications (1)

Publication Number Publication Date
WO2017016240A1 true WO2017016240A1 (zh) 2017-02-02

Family

ID=55100779

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/078869 WO2017016240A1 (zh) 2015-07-24 2016-04-08 一种钞票冠字号识别方法

Country Status (2)

Country Link
CN (1) CN105261109B (zh)
WO (1) WO2017016240A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111539414A (zh) * 2020-04-26 2020-08-14 梁华智能科技(上海)有限公司 一种ocr图像字符识别和字符校正的方法及***
CN111709419A (zh) * 2020-06-10 2020-09-25 中国工商银行股份有限公司 一种纸币冠字号的定位方法、***、设备及可读存储介质
CN113221801A (zh) * 2021-05-24 2021-08-06 北京奇艺世纪科技有限公司 版号信息识别方法、装置、电子设备及可读存储介质
CN114140928A (zh) * 2021-11-19 2022-03-04 苏州益多多信息科技有限公司 一种高精准度的数字彩统一化查票方法、***及介质
CN117671849A (zh) * 2023-12-14 2024-03-08 浙江南星科技有限公司 一种采用滑钞结构的立式图像扫描点钞机

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105261109B (zh) * 2015-07-24 2017-10-31 广州广电运通金融电子股份有限公司 一种钞票冠字号识别方法
CN105913547B (zh) * 2016-04-07 2018-11-20 四川大学 一种账票光学识别方法及装置
CN108154596B (zh) * 2016-12-04 2020-11-10 湖南丰汇银佳科技股份有限公司 一种基于图像匹配的双冠号纸币鉴伪方法
CN106887077B (zh) * 2017-01-17 2019-07-09 深圳怡化电脑股份有限公司 纸币交易信息查询方法及装置
CN108320373B (zh) * 2017-01-17 2020-07-24 深圳怡化电脑股份有限公司 一种纸币防伪标识的检测的方法及装置
CN108806058A (zh) * 2017-05-05 2018-11-13 深圳怡化电脑股份有限公司 一种纸币检测方法及装置
CN107240185B (zh) * 2017-06-23 2019-09-20 深圳怡化电脑股份有限公司 一种冠字号识别方法、装置、设备及存储介质
CN107358718B (zh) * 2017-07-10 2019-09-20 深圳怡化电脑股份有限公司 一种冠字号识别方法、装置、设备及存储介质
CN109635796B (zh) * 2018-11-20 2021-09-28 泰康保险集团股份有限公司 调查问卷的识别方法、装置和设备
CN109871847B (zh) * 2019-03-13 2022-09-30 厦门商集网络科技有限责任公司 一种ocr识别方法及终端
CN111783766B (zh) * 2020-07-10 2023-02-14 上海淇毓信息科技有限公司 一种分步识别图像字符的方法、装置和电子设备
CN113963359B (zh) * 2021-12-20 2022-03-18 北京易真学思教育科技有限公司 文本识别模型训练方法、文本识别方法、装置及电子设备
CN115035313B (zh) * 2022-06-15 2023-01-03 云南这里信息技术有限公司 黑颈鹤识别方法、装置、设备及存储介质
CN117197828A (zh) * 2023-08-11 2023-12-08 中国银行保险信息技术管理有限公司 票据信息识别方法、装置、介质及设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007105892A1 (en) * 2006-03-13 2007-09-20 Nautilus Hyosung Inc. Recognizing the denomination of a note using wavelet transform
CN101944174A (zh) * 2009-07-08 2011-01-12 西安电子科技大学 车牌字符的识别方法
CN102663377A (zh) * 2012-03-15 2012-09-12 华中科技大学 一种基于模板匹配的字符识别方法
CN104156701A (zh) * 2014-07-26 2014-11-19 佳都新太科技股份有限公司 一种基于决策树和svm的车牌相似字符识别方法
CN104318238A (zh) * 2014-11-10 2015-01-28 广州御银科技股份有限公司 一种验钞模块中对扫描的钞票图提取冠字号的方法
CN105261109A (zh) * 2015-07-24 2016-01-20 广州广电运通金融电子股份有限公司 一种钞票冠字号识别方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007105892A1 (en) * 2006-03-13 2007-09-20 Nautilus Hyosung Inc. Recognizing the denomination of a note using wavelet transform
CN101944174A (zh) * 2009-07-08 2011-01-12 西安电子科技大学 车牌字符的识别方法
CN102663377A (zh) * 2012-03-15 2012-09-12 华中科技大学 一种基于模板匹配的字符识别方法
CN104156701A (zh) * 2014-07-26 2014-11-19 佳都新太科技股份有限公司 一种基于决策树和svm的车牌相似字符识别方法
CN104318238A (zh) * 2014-11-10 2015-01-28 广州御银科技股份有限公司 一种验钞模块中对扫描的钞票图提取冠字号的方法
CN105261109A (zh) * 2015-07-24 2016-01-20 广州广电运通金融电子股份有限公司 一种钞票冠字号识别方法

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111539414A (zh) * 2020-04-26 2020-08-14 梁华智能科技(上海)有限公司 一种ocr图像字符识别和字符校正的方法及***
CN111539414B (zh) * 2020-04-26 2023-05-23 梁华智能科技(上海)有限公司 一种ocr图像字符识别和字符校正的方法及***
CN111709419A (zh) * 2020-06-10 2020-09-25 中国工商银行股份有限公司 一种纸币冠字号的定位方法、***、设备及可读存储介质
CN113221801A (zh) * 2021-05-24 2021-08-06 北京奇艺世纪科技有限公司 版号信息识别方法、装置、电子设备及可读存储介质
CN113221801B (zh) * 2021-05-24 2023-08-18 北京奇艺世纪科技有限公司 版号信息识别方法、装置、电子设备及可读存储介质
CN114140928A (zh) * 2021-11-19 2022-03-04 苏州益多多信息科技有限公司 一种高精准度的数字彩统一化查票方法、***及介质
CN114140928B (zh) * 2021-11-19 2023-08-22 苏州益多多信息科技有限公司 一种高精准度的数字彩统一化查票方法、***及介质
CN117671849A (zh) * 2023-12-14 2024-03-08 浙江南星科技有限公司 一种采用滑钞结构的立式图像扫描点钞机
CN117671849B (zh) * 2023-12-14 2024-05-14 浙江南星科技有限公司 一种采用滑钞结构的立式图像扫描点钞机

Also Published As

Publication number Publication date
CN105261109B (zh) 2017-10-31
CN105261109A (zh) 2016-01-20

Similar Documents

Publication Publication Date Title
WO2017016240A1 (zh) 一种钞票冠字号识别方法
Shahab et al. ICDAR 2011 robust reading competition challenge 2: Reading text in scene images
EP2808827B1 (en) System and method for OCR output verification
TW389865B (en) System and method for automated interpretation of input expressions using novel a posteriori probability measures and optimally trained information processing network
WO2020164278A1 (zh) 一种图像处理方法、装置、电子设备和可读存储介质
CN106845358B (zh) 一种手写体字符图像特征识别的方法及***
CN106203539B (zh) 识别集装箱箱号的方法和装置
Zawbaa et al. An automatic flower classification approach using machine learning algorithms
CN108108760A (zh) 一种快速人脸识别方法
CN108681735A (zh) 基于卷积神经网络深度学习模型的光学字符识别方法
Pirrone et al. Papy-s-net: A siamese network to match papyrus fragments
CN110826408A (zh) 一种分区域特征提取人脸识别方法
US9002115B2 (en) Dictionary data registration apparatus for image recognition, method therefor, and program
Pan et al. Improving scene text detection by scale-adaptive segmentation and weighted CRF verification
Liu et al. Gender identification in unconstrained scenarios using self-similarity of gradients features
CN111213157A (zh) 一种基于智能终端的快递信息录入方法及录入***
CN107103289A (zh) 利用笔迹轮廓特征来进行笔迹鉴别的方法及***
Işikdoğan et al. Automatic recognition of Turkish fingerspelling
Halder et al. Individuality of isolated Bangla characters
CN111612045B (zh) 一种获取目标检测数据集的通用方法
Gyamfi et al. Pixel-based unsupervised classification approach for information detection on optical markup recognition sheet
Rajithkumar et al. Template matching method for recognition of stone inscripted Kannada characters of different time frames based on correlation analysis
CN115272689A (zh) 基于视图的空间形状识别方法、装置、设备和存储介质
Athoillah et al. Handwritten arabic numeral character recognition using multi kernel support vector machine
CN113111882B (zh) 一种卡证识别方法、装置、电子设备及存储介质

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: 16829607

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: 16829607

Country of ref document: EP

Kind code of ref document: A1