WO2014026483A1 - Procédé d'identification de caractères et dispositif approprié - Google Patents

Procédé d'identification de caractères et dispositif approprié Download PDF

Info

Publication number
WO2014026483A1
WO2014026483A1 PCT/CN2013/073927 CN2013073927W WO2014026483A1 WO 2014026483 A1 WO2014026483 A1 WO 2014026483A1 CN 2013073927 W CN2013073927 W CN 2013073927W WO 2014026483 A1 WO2014026483 A1 WO 2014026483A1
Authority
WO
WIPO (PCT)
Prior art keywords
character
boundary
pixel
center
boundary direction
Prior art date
Application number
PCT/CN2013/073927
Other languages
English (en)
Chinese (zh)
Inventor
向拓闻
关玉萍
徐朝阳
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 WO2014026483A1 publication Critical patent/WO2014026483A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/14—Image acquisition
    • G06V30/148—Segmentation of character regions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/18—Extraction of features or characteristics of the image
    • G06V30/182—Extraction of features or characteristics of the image by coding the contour of the pattern
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/18—Extraction of features or characteristics of the image
    • G06V30/182—Extraction of features or characteristics of the image by coding the contour of the pattern
    • G06V30/1823—Extraction of features or characteristics of the image by coding the contour of the pattern using vector-coding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition

Definitions

  • the present invention relates to the field of image processing, and in particular, to a character recognition method and related apparatus.
  • Character recognition refers to the process of scanning text data, then analyzing and processing the image file to obtain text and layout information.
  • character recognition technology has gradually entered various fields such as people's life, study and work. Such as: price file serial number identification, license plate recognition, etc.
  • the character recognition methods mainly include: template matching method, background feature method, and the like.
  • the template matching algorithm is simple and easy to operate, but the characters are standardized, the size is the same, the template has a strict fixed position, and the template matching is computationally intensive and time-consuming;
  • the feature method recognizes the speed faster, but the position of the anchor point is high, depending on the stability of the feature point position.
  • Embodiments of the present invention provide a character recognition method and related apparatus for quickly identifying characters.
  • the character recognition method provided by the present invention includes: acquiring image data of a character; determining a background pixel point and a foreground pixel point in the image data; locating a center of gravity of the foreground pixel point, and dividing the center of gravity into a segmentation center point a character region; determining a boundary direction of a boundary pixel in each of the character regions; separately counting boundary directions of all boundary pixel points in each of the character regions, and obtaining boundary direction features of each of the character regions; The boundary direction feature within the region identifies the character.
  • the determining the background pixel and the foreground pixel in the image data includes: comparing a background threshold and a pixel value of each pixel in the image data, respectively, if the pixel value is less than the background threshold, The pixel corresponding to the pixel value is the foreground pixel; if the pixel value If the background threshold is greater than or equal to, the pixel corresponding to the pixel value is a background pixel.
  • the method before the determining the background pixel point and the foreground pixel point in the image data, includes: preprocessing the image data, where the preprocessing includes any one of noise filtering, tilt correction, and character segmentation. Or a combination of two or more.
  • the center of gravity of the positioning foreground pixel point includes:
  • the determining a boundary direction of a boundary pixel point in each of the character regions includes: calculating, according to a boundary pixel point, a lateral offset of the boundary pixel point according to a neighborhood direction of the boundary pixel point a derivative and a longitudinal partial derivative; determining a boundary direction of the boundary pixel point in combination with the lateral partial derivative and the value of the longitudinal partial derivative.
  • the boundary direction of all the boundary pixels in each of the character regions is separately calculated, and the boundary direction features of each of the character regions are obtained, including:
  • the boundary direction of all boundary pixels in a character region is Counting the boundary direction of all boundary pixels in a character region; using a combination of vector values in each boundary direction as a feature vector of the one character region, the feature vector of the character region is a boundary direction feature of the character region; The vector value of the boundary direction is the proportion of the pixel points in one boundary direction in all the pixel points in the one character region.
  • the recognizing the character according to the boundary direction feature in each of the character regions includes: performing a comparison, respectively calculating a variance, and using a character corresponding to the minimum variance as an output result of the character recognition.
  • step 2) Using the method of step 1) to sample the boundary direction features of the different types of fonts of the character, and by using the mean value of the boundary direction features of various types of fonts of the same character, the mean value is used as the standard boundary direction feature of the character.
  • the number of samples of the character is Nx, and the Nx is less than 10000 and greater than 10.
  • the determining the background pixel and the foreground pixel in the image data includes: comparing a background threshold and a pixel value of each pixel in the image data, respectively, if the pixel value is less than the background threshold, The pixel point corresponding to the pixel value is a foreground pixel point; if the pixel value is greater than or equal to the background threshold, the pixel point corresponding to the pixel value is a background pixel point.
  • the method before the determining the background pixel point and the foreground pixel point in the image data, the method includes: preprocessing the image data, where the preprocessing includes any one of noise filtering, tilt correction, and character segmentation. Or a combination of two or more.
  • the center of gravity of the positioning foreground pixel point includes:
  • the determining a boundary direction of a boundary pixel point in each of the character regions includes: calculating, according to a boundary pixel point, a lateral offset of the boundary pixel point according to a neighborhood direction of the boundary pixel point a derivative and a longitudinal partial derivative; determining a boundary direction of the boundary pixel point in combination with the lateral partial derivative and the value of the longitudinal partial derivative.
  • the boundary direction of all the boundary pixels in each of the character regions is separately calculated, and the boundary direction features of each of the character regions are obtained, including:
  • the boundary direction of all boundary pixels in a character region is Counting the boundary direction of all boundary pixels in a character region; using a combination of vector values in each boundary direction as a feature vector of the one character region, the feature vector of the character region is a boundary direction feature of the character region; The vector value of the boundary direction is the proportion of the pixel points in one boundary direction in all the pixel points in the one character region.
  • An image obtaining unit configured to acquire image data of a character
  • a pixel identification unit configured to determine a background pixel and a foreground pixel in the image data
  • a region dividing unit configured to locate a center of gravity of the foreground pixel, and divide the character region by using the center of gravity as a segmentation center point
  • a direction determining unit configured to determine a boundary direction of a boundary pixel in each of the character regions, and a feature statistic unit, configured to separately calculate boundary directions of all boundary pixel points in each of the character regions, to obtain each of the character regions Boundary direction feature;
  • a character recognition unit configured to identify the character according to a boundary direction feature in each of the character regions.
  • the device further includes:
  • a pre-processing unit configured to perform pre-processing on the image data, where the pre-processing includes any one or a combination of two or more of noise filtering, tilt correction, and character segmentation.
  • the direction determining unit includes:
  • a partial derivative calculation module configured to calculate a lateral partial derivative and a longitudinal partial derivative of the boundary pixel according to a neighborhood direction of the boundary pixel, centering on a boundary pixel;
  • a direction identifying module configured to determine a boundary direction of the boundary pixel points in combination with the values of the lateral partial derivative and the longitudinal partial derivative.
  • the feature statistics unit includes:
  • a directional statistics module configured to calculate a boundary direction of all boundary pixels in a character region
  • a vector combination module configured to use a combination of vector values in each boundary direction as a feature vector of the one character region, where the character region
  • the feature vector is a boundary direction feature of the character region
  • the vector value of the boundary direction is a proportion of a pixel point of one boundary direction in all pixel points in the one character region.
  • Identifying a sample unit for acquiring image data of a character determining a background pixel point and a foreground pixel point in the image data; locating a center of gravity of the foreground pixel point, and dividing the character area by the center of gravity as a segmentation center point; Determining a boundary direction of a boundary pixel in each of the character regions; separately counting boundary directions of all boundary pixel points in each of the character regions, and obtaining and storing boundary direction features of each of the character regions of the character;
  • the directional feature counts the mean of the boundary direction features of the various types of fonts of the same character, and uses the mean as the standard boundary direction feature of the character.
  • the center of gravity of the positioning foreground pixel is specifically:
  • the determining a boundary direction of a boundary pixel in each of the character regions is specifically:
  • boundary direction of all boundary pixel points in each of the character regions is separately calculated, and the boundary direction features of each of the character regions are obtained, specifically:
  • the boundary direction of all boundary pixels in a character region is Counting the boundary direction of all boundary pixels in a character region; using a combination of vector values in each boundary direction as a feature vector of the one character region, the feature vector of the character region is a boundary direction feature of the character region; The vector value of the boundary direction is the proportion of the pixel points in one boundary direction in all the pixel points in the one character region.
  • the embodiments of the present invention have the following advantages:
  • feature extraction is performed on a character in a region, which can effectively avoid the overall feature deviation caused by noise interference in a certain region, and the character region is divided into a center point of the foreground pixel point of the character, which is effective.
  • the embodiment of the invention defines the boundary direction of the characters, and different feature vectors can be extracted according to different boundary combinations of the characters according to the boundary features of the characters, and different types of fonts are extracted. Adaptable.
  • FIG. 1 is a schematic flow chart of a character recognition method according to an embodiment of the present invention.
  • FIG. 2 is another schematic flowchart of a character recognition method according to an embodiment of the present invention.
  • 3 is a schematic diagram of division of a center of gravity of a character in an embodiment of the present invention
  • 4 is a schematic diagram of a boundary direction in an embodiment of the present invention
  • Figure 5 is another schematic view of the boundary direction in the embodiment of the present invention.
  • FIG. 6 is a schematic diagram of comparison of center of gravity and center division in an embodiment of the present invention.
  • FIG. 7 is a schematic diagram of a sample statistics in an embodiment of the present invention.
  • FIG. 8 is a schematic diagram showing the logical structure of a character recognition apparatus according to an embodiment of the present invention.
  • FIG. 9 is a schematic diagram showing the logical structure of a character sampling device according to an embodiment of the present invention.
  • Embodiments of the present invention provide a character recognition method and related apparatus for quickly identifying characters.
  • an embodiment of a character recognition method in an embodiment of the present invention includes:
  • the character recognition device collects image data of characters through a contact image sensor (CIS, Contact Image Sensor), a camera or other collection device, the image data including pixel points, and gray value data of the pixel points.
  • CIS Contact Image Sensor
  • a camera or other collection device the image data including pixel points, and gray value data of the pixel points.
  • the character recognition means determines a background pixel and a foreground pixel in the image data, the foreground pixel is a pixel represented by a character, and the background pixel is a pixel represented by a character background.
  • the character recognition device locates a center of gravity of the foreground pixel, and divides a character region by using the center of gravity as a segmentation center point; the center of gravity is a coordinate center point of the foreground pixel point.
  • some recognition techniques perform feature extraction on an image as a whole. Once a noise point occurs in a certain area of a character (for example, a character is stained), an error occurs in the recognition of the entire character, and the present invention implements For example, sub-region processing of characters and extracting and identifying features of each region separately can effectively reduce errors caused by local noise.
  • the characters can be divided into four areas, eight areas or sixteen areas, which are determined according to the accuracy of the identification, which is not limited this time.
  • the character recognition means determines a boundary direction of boundary pixel points in each of said character regions; said boundary direction is a normal vector direction of a tangent to each boundary point.
  • the boundary directions of all the boundary pixels in each of the character regions are respectively counted; the character recognition device separately collects boundary directions of all boundary pixel points in each of the character regions, and obtains boundary direction features of each of the character regions;
  • the boundary direction feature embodies the boundary characteristics of the boundary pixel.
  • the character recognition means recognizes the word payment based on the boundary direction feature in each of the character areas.
  • the character recognition device presets a standard value of the boundary direction feature of each character, and the character recognition device can complete the recognition of the character by comparing or matching the boundary direction feature of the character to be recognized with the standard value.
  • feature extraction is performed on a character in a region, which can effectively avoid the overall feature deviation caused by noise interference in a certain region, and the character region is divided into a center point of the foreground pixel point of the character, which is effective.
  • the embodiment of the invention defines the boundary direction of the characters, and different feature vectors can be extracted according to different boundary combinations of the characters according to the boundary features of the characters, and different types of fonts are extracted. Adaptable.
  • FIG. 2 another embodiment of the character recognition method in the embodiment of the present invention includes:
  • the character recognition device collects image data of characters through a CIS, a camera or other collection device, the image data including pixel points, and gray value data of the pixel points.
  • the character recognition means preprocesses the image data, and the preprocessing includes any one or a combination of two or more of noise filtering, tilt correction, and character division.
  • the image data of the character will generate noise during the collection process, and the character image noise is first filtered by Gaussian filtering.
  • the character recognition device extracts an edge point of the image data, and performs the edge point Straight line fitting, obtaining the inclination angle of the edge point after the straight line fitting, and finally adjusting the image data according to the inclination angle such that the upper and lower boundaries of the image data are parallel to the horizontal plane.
  • the character recognition device compares the background threshold and the pixel value of each pixel in the image data, and if the pixel value is smaller than the background threshold, the pixel corresponding to the pixel value is a foreground pixel; If the background threshold is greater than or equal to, the pixel corresponding to the pixel value is a background pixel.
  • the image data is binarized, for example, the background pixel is set to 0 and the foreground pixel is set to 1.
  • the character recognition device locates the center of gravity of the foreground pixel; specifically, the character recognition device separately counts the sum of the lateral coordinate values of the respective pixels in the foreground pixel and the sum of the longitudinal coordinate values, and divides the sum of the horizontal coordinate values by The value of the total number of foreground pixel points is taken as the abscissa of the center of gravity, and the value of the sum of the longitudinal coordinate values divided by the total number of the foreground pixel points is taken as the ordinate of the center of gravity.
  • the character area is divided by the center of gravity as a segmentation center point
  • the character recognition device locates the character region by dividing the center of gravity into a center point; optionally, the character can be divided into four regions, eight regions or sixteen regions, which are determined according to the accuracy of the recognition, which is not limited this time. .
  • the character recognition means determines a boundary direction of boundary pixel points in each of the character regions; the boundary direction is a normal vector direction of a tangent to each boundary point.
  • the method for determining the boundary direction of the boundary pixel may be: searching for a boundary pixel in the target character region, centering on the boundary pixel, and calculating a lateral partial derivative of the boundary pixel according to a neighborhood direction of the boundary pixel And a longitudinal partial derivative; determining a boundary direction of the boundary pixel point in combination with the lateral partial derivative and the value of the longitudinal partial derivative.
  • the boundary directions of all boundary pixel points in each of the character regions are respectively counted.
  • the character recognition device separately counts boundary directions of all boundary pixel points in each of the character regions, and obtains boundary direction features of each of the character regions.
  • the method for obtaining the boundary direction feature of the character region may be: counting the boundary direction of all boundary pixel points in a character region, and combining the vector values in the respective boundary directions as the feature vector of the one character region, where the character region is
  • the feature vector is a boundary direction feature of the character region; the vector value of the boundary direction is a proportion of a pixel point of one boundary direction in all pixel points in the one character region.
  • there are eight predefined boundary directions and the number of the eight directions in the target character area is (1, 7, 1 , 0, 1 , 3 , 0, 0), and after normalization, the features are obtained.
  • the vector is ( 0.076923 , 0.538461 , 0.076923 , 0.000000 , 0.076923 , 0.230769 , 0.000000 , 0.000000 , 0.000000 )
  • the character recognition means recognizes the word payment based on the boundary direction feature in each of the character areas.
  • the standard boundary direction features of all characters are compared, and the variance is calculated separately, and the character corresponding to the minimum variance is used as the output result of the character recognition.
  • the character sampling device collects a single-character image of various fonts through a CIS tube, a camera or other collection device, and the single-character image contains only one character image, and the collected character contains the character to be recognized, this embodiment
  • the number characters '0'-'9' are included.
  • the character sampling device binarizes the single-character image to obtain a single-character binary image with the background pixel point being 0 and the foreground pixel point being 1.
  • the j-th column pixel value, 0 ('', is the i-th row of the output image 0 the j-th column pixel value, ⁇ ./ ⁇ is the background threshold.
  • the character sampling device positions the center of gravity of the single-character binarized image, wherein the lower left corner of the single-character binarized image can be set as the coordinate origin (0, 0), and the single-character binarized image is divided according to the center of gravity
  • the boundary direction features of each region are calculated as features of the characters.
  • the center of gravity of the image '4' is positioned, wherein the lower left corner of the image is the origin, from left to right, from top to bottom, and progressively, if it is a character region, the horizontal coordinate value xCount is respectively accumulated.
  • the ordinate value yCount while accumulating the number of pixels in the character area Ow"t, traversing the total number of pixels in the full range, that is, you can get
  • the character sampling device performs feature partitioning.
  • feature extraction is not performed on the entire image, but image extraction is performed on the image sub-region, so as to avoid deviation of the overall feature when noise is disturbed in a certain region.
  • the specific representation method is: Calculate the partial derivative of the four neighborhood directions, and the lateral partial derivative, centering on a certain point:
  • the character sampling device calculates eight directions in the upper left corner region (region 1) of FIG. 3, and the search method is: from left to right, from top to bottom, if the point is a background point (pixel value is 0), Then determine whether it is a boundary point, determine whether the four neighboring points have a character area (pixel value is 1), if it is, the point is considered to be a boundary point, otherwise it is not a boundary point, as shown in Figure 3, point 1 its four neighbors If the domain is the background point 0, then it is not the boundary point.
  • the point 2 has a point of 1 in the four neighborhoods, which is the character area, and the point 2 is the boundary point.
  • the direction of the boundary point is determined according to the information of the four neighborhoods; the four neighborhoods of point 3 conform to the definition of b-2 of FIG. 5, then the direction of the point is 2; the four neighborhoods of point 4 conform to e-5 of FIG.
  • the direction of the point is 5.
  • the contrast effect of the two methods is shown in 6. It can also be seen from the image that the image is partitioned by the center of gravity, and the image is offset with good consistency. 8a and 8b are the case where the image character area of the center of gravity is centered and offset, and 8c and 8d are the case where the character area of the image center sub-area is centered and offset.
  • the character sampling device counts the standard boundary direction features
  • the feature vector of each numeric character can be obtained, including 32 features.
  • the sample number of each font is Nx (10 ⁇ Nx ⁇ 10000), as the character standard.
  • the boundary direction feature is pre-stored in the storage unit. The number of boundary points in eight directions, and the feature vector where ' ⁇ 1, 2 , 3 , 4 ⁇ .
  • the feature vector of the character is D1, T2, T3, T4 ⁇ , and the statistical N (100 ⁇ N ⁇ 10000) samples, where N is the average value of each of its eigenvectors.
  • Direction feature feature (feal,
  • the character recognition device collects a digital character image to be recognized by a CIS tube, a camera or other collection device, the character image containing one or more numeric characters.
  • the character recognition device performs preprocessing of the image data. Due to the collection device, the character image generates noise during the collection process. First, the character image noise is filtered by Gaussian filtering; the character image is corrected, and the tilt problem of the character image is processed. Then, the image is subjected to threshold processing, and there are many threshold values.
  • the present invention selects the maximum variance threshold method, and normalizes the binarized image so that the character region is represented by 1, and the background region is represented by 0;
  • ('', ) is the i-th row of the input original image, the pixel value of the j-th column, 0 ('', is the i-th row of the output image 0, the pixel value of the j-th column, re ⁇ o is the background threshold;
  • the character recognition device then divides the single-character binary image by a priori knowledge and horizontal and vertical projection methods, and the single-character binary image contains only one numeric character.
  • the character recognition device performs character feature extraction, character n feature extraction, and locates the character n binarized image center of gravity, and specifies the lower left corner of the image as the coordinate origin (0, 0), from left to right, from top to bottom, Line scan, if it is a character area, accumulate its abscissa value xCo nt and ordinate value yc respectively. ""t, at the same time accumulate the number of pixels in the character area Ow"t, after traversing the entire image, the total number of pixels in the horizontal area, that is, the character area can be obtained
  • xCenterofGrav xCount I Count
  • yCenterofGrav yCount I Count
  • the character recognition device performs feature partitioning; taking the character n binarized image center of gravity as the center, the character! !
  • the number of eight directions in the first region is (2, 0, 7, 0, 0, 8, 0), and the number of directions in the two regions is (0, 1, 7, respectively).
  • 0, 2, 2, 0, 1) the number of the eight directions in the three regions is (5, 2, 1, 0, 1, 2, 2, 0), and the number of the eight directions in the region is (0).
  • the character recognition device performs feature comparison; by the feature that the character n is currently to be recognized
  • the character recognition device calculates the variance of the character to be recognized and the ten standard features, respectively, as shown in the following table:
  • An embodiment of the character recognition apparatus in the embodiment of the present invention includes:
  • a pixel identifying unit 802 configured to determine a background pixel point and a foreground pixel point in the image data
  • a region dividing unit 803 configured to locate a center of gravity of the foreground pixel point, and divide the character area by using the center of gravity as a dividing center point
  • the direction determining unit 804 is configured to determine a boundary direction of a boundary pixel point in each character region, and a feature statistic unit 805, configured to separately calculate boundary directions of all boundary pixel points in each of the character regions, to obtain each of the character regions.
  • Boundary direction feature
  • the character recognition unit 806 is configured to recognize the character according to a boundary direction feature in each of the character regions.
  • the apparatus in the embodiment of the present invention may further include:
  • the pre-processing unit 807 is configured to perform pre-processing on the image data, and the pre-processing includes any one or a combination of two or more of noise filtering, tilt correction, and character segmentation.
  • the direction determining unit 804 in the embodiment of the present invention may further include: a partial derivative calculation module 8041, configured to calculate the boundary according to a neighborhood direction of the boundary pixel by using a boundary pixel as a center The lateral partial derivative and the longitudinal partial derivative of the pixel;
  • the direction identifying module 8042 is configured to determine a boundary direction of the boundary pixel by combining the values of the lateral partial derivative and the longitudinal partial derivative.
  • the feature statistics unit 805 in the embodiment of the present invention includes:
  • the direction statistics module 8051 is configured to count boundary directions of boundary pixels in a character region.
  • a vector combination module 8052 configured to use a combination of vector values in respective boundary directions as a feature vector of the one character region, where a feature vector of the character region is an edge of the character region
  • the boundary direction feature the vector value of the boundary direction is the proportion of the pixel points in one boundary direction in all the pixels in the one character region.
  • the specific interaction process of each unit in the character recognition apparatus in the embodiment of the present invention is as follows:
  • the image acquisition unit 801 collects image data of characters through a CIS, a camera or other collection device, the image data includes pixel points, and gray levels of pixel points. Value data.
  • the pre-processing unit 807 preprocesses the image data, and the pre-processing includes any one or a combination of two or more of noise filtering, tilt correction, and character division.
  • the image data of the character will generate noise during the collection process, and the character image noise is first filtered by Gaussian filtering.
  • the character recognition device extracts an edge point of the image data, and performs a straight line on the edge point. After fitting, the inclination angle of the edge point after the straight line fitting is acquired, and finally the image data is adjusted according to the inclination angle such that the upper and lower boundaries of the image data are parallel to the horizontal plane.
  • the pixel identification unit 802 compares the background threshold and the pixel value of each pixel in the image data respectively. If the pixel value is smaller than the background threshold, the pixel corresponding to the pixel value is a foreground pixel; if the pixel value is greater than or Equal to the background threshold, the pixel corresponding to the pixel value is a background pixel. After the background pixel and the foreground pixel in the image data are determined, the image data is binarized, for example, the background pixel is set to 0 and the foreground pixel is set to 1.
  • the area dividing unit 803 locates the center of gravity of the foreground pixel point. Specifically, the character recognition device separately calculates the sum of the horizontal coordinate values of the respective pixel points in the foreground pixel point and the sum of the longitudinal coordinate values, and divides the sum of the horizontal coordinate values. Taking the value of the total number of foreground pixel points as the abscissa of the center of gravity, the value of the sum of the longitudinal coordinate values divided by the total number of the foreground pixel points as the ordinate of the center of gravity; the area dividing unit 803 The center of gravity divides the character area for the segmentation center point.
  • the character can be divided into four regions, eight regions or sixteen regions, which need to be determined according to the accuracy of the recognition, which is not limited this time.
  • the direction determining unit 804 determines a boundary direction of boundary pixel points in each of the character regions; the boundary direction is a normal vector direction of a tangent to each boundary point.
  • the method for determining the boundary direction of the boundary pixel may be: searching for a boundary pixel in the target character region, centering on the boundary pixel, and calculating a lateral partial derivative of the boundary pixel according to a neighborhood direction of the boundary pixel And a longitudinal partial derivative; determining a boundary direction of the boundary pixel point in combination with the lateral partial derivative and the value of the longitudinal partial derivative.
  • the feature statistic unit 805 separately counts the boundary directions of all the boundary pixels in each of the character regions, and obtains the boundary direction features of each of the character regions.
  • the method for obtaining the boundary direction feature of the character region may be: counting the boundary direction of all boundary pixel points in a character region, and combining the vector values in the respective boundary directions as the feature vector of the one character region, where the character region is The feature vector is a boundary direction feature of the character region; the vector value of the boundary direction is a proportion of a pixel point of one boundary direction in all pixel points in the one character region. For example, there are eight predefined boundary directions, and the number of the eight directions in the target character area is (1, 7, 1, 0, 1, 3, 0, 0), and after normalization, the features are obtained.
  • the vector is ( 0.076923 , 0.538461 , 0.076923 , 0.000000 , 0.076923 , 0.230769 , 0.000000 , 0.000000 ) naval character.
  • the character recognition device can perform the boundary direction feature of the character and the standard boundary direction feature of all characters in the font respectively.
  • the variance is calculated separately, and the character corresponding to the minimum variance is used as the output result of the character recognition.
  • An embodiment of the character sampling device of the present invention for performing the above character recognition method.
  • An embodiment of the character sampling device in the embodiment of the present invention includes:
  • Identifying a sample unit 901, configured to acquire image data of a character; determining a background pixel point and a foreground pixel point in the image data; locating a center of gravity of the foreground pixel point, and dividing the character area by using the center of gravity as a segmentation center point Determining a boundary direction of a boundary pixel in each of the character regions; separately counting boundary directions of all boundary pixel points in each of the character regions, and obtaining a standard sample unit 902 for using the identification sample unit
  • the boundary direction feature of the target character is counted, and the mean value of the boundary direction feature of each type of font of the same character is counted, and the mean value is used as a standard boundary direction feature of the character.
  • the center of gravity of the positioning foreground pixel is specifically:
  • Determining a boundary direction of a boundary pixel in each of the character regions specifically: calculating, according to a boundary pixel point, a lateral partial derivative and a vertical direction of the boundary pixel according to a neighborhood direction of the boundary pixel Partial derivative; determining a boundary direction of the boundary pixel point in combination with the value of the lateral partial derivative and the longitudinal partial derivative.
  • boundary direction of all the boundary pixels in each of the character regions is separately calculated, and the boundary direction features of each of the character regions are obtained, which are specifically:
  • the boundary direction of all boundary pixels in a character region is Counting the boundary direction of all boundary pixels in a character region; using a combination of vector values in each boundary direction as a feature vector of the one character region, the feature vector of the character region is a boundary direction feature of the character region; The vector value of the boundary direction is the proportion of the pixel points in one boundary direction in all the pixel points in the one character region.
  • the disclosed apparatus and method can be implemented in other ways.
  • 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 separate, and the components displayed as units may or may not be physical units, i.e., may be located in one place, or may be distributed over 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 of the functional units in the various embodiments of the present invention may be integrated into one processing unit.
  • each unit may exist physically separately, or two or more units may be integrated into one unit.
  • 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 may contribute 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, which can store program codes. .

Landscapes

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

Abstract

L'invention concerne un procédé d'identification de caractères et un dispositif approprié, utilisés pour identifier des caractères de manière efficiente et rapide. Selon le mode de réalisation de la présente invention, le procédé comporte les étapes consistant à : obtenir des données d'image d'un caractère; déterminer un point de pixel d'arrière-plan et un point de pixel d'avant-plan dans les données d'image; positionner le centre de gravité du point de pixel d'avant-plan et diviser des régions de caractère en utilisant le centre de gravité comme point central de découpe; déterminer une direction de bordure d'un point de pixel de bordure dans chaque région de caractère; compter respectivement les directions de bordure de tous les points de pixels de bordure dans chaque région de caractère pour obtenir une caractéristique de direction de bordure de chaque région de caractère; et identifier le caractère en fonction de la caractéristique de direction de bordure dans chaque région de caractère.
PCT/CN2013/073927 2012-08-15 2013-04-09 Procédé d'identification de caractères et dispositif approprié WO2014026483A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201210290700.X 2012-08-15
CN201210290700XA CN102831416A (zh) 2012-08-15 2012-08-15 一种字符识别方法及相关装置

Publications (1)

Publication Number Publication Date
WO2014026483A1 true WO2014026483A1 (fr) 2014-02-20

Family

ID=47334544

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2013/073927 WO2014026483A1 (fr) 2012-08-15 2013-04-09 Procédé d'identification de caractères et dispositif approprié

Country Status (2)

Country Link
CN (1) CN102831416A (fr)
WO (1) WO2014026483A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115601376A (zh) * 2022-09-22 2023-01-13 北京华益精点生物技术有限公司（Cn） 图像裁切方法、装置、电子设备及存储介质

Families Citing this family (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102831416A (zh) * 2012-08-15 2012-12-19 广州广电运通金融电子股份有限公司 一种字符识别方法及相关装置
CN103049750B (zh) * 2013-01-11 2016-06-15 广州广电运通金融电子股份有限公司 字符识别方法
CN103544491A (zh) * 2013-11-08 2014-01-29 广州广电运通金融电子股份有限公司 一种面向复杂背景的光学字符识别方法及装置
CN104361312B (zh) * 2014-10-16 2017-11-14 北京捷通华声语音技术有限公司 一种对图像进行字符识别的方法和装置
CN104463126A (zh) * 2014-12-15 2015-03-25 湖南工业大学 一种扫描文档图像的倾斜角自动检测方法
CN104777931A (zh) * 2015-03-24 2015-07-15 深圳市艾优尼科技有限公司 一种终端
CN106340001B (zh) * 2015-07-07 2019-02-15 富士通株式会社 图像划分装置和图像划分方法
CN105335745B (zh) * 2015-11-27 2018-12-18 小米科技有限责任公司 图像中数字的识别方法、装置及设备
CN105488504A (zh) * 2015-12-28 2016-04-13 四川长虹网络科技有限责任公司 基于摄像头的汉字识别方法
CN105741561B (zh) * 2016-03-11 2018-03-16 中国联合网络通信集团有限公司 车牌特征位的识别方法及装置
CN107784301B (zh) * 2016-08-31 2021-06-11 百度在线网络技术（北京）有限公司 用于识别图像中文字区域的方法和装置
CN106408266A (zh) * 2016-09-29 2017-02-15 广州鹤互联网科技有限公司 一种签核文件自动生成方法及装置
CN106682666A (zh) * 2016-12-29 2017-05-17 成都数联铭品科技有限公司 用于非常见字体ocr识别的特征模板制备方法
CN111079735B (zh) * 2018-10-22 2023-12-22 莱芜钢铁集团电子有限公司 一种铸坯端面字符识别方法及装置
CN109753967A (zh) * 2018-12-29 2019-05-14 北京师范大学 一种图片文字识别方法
CN109871521A (zh) * 2019-01-08 2019-06-11 平安科技（深圳）有限公司 一种电子文档的生成方法及设备
CN111027546B (zh) * 2019-12-05 2024-03-26 嘉楠明芯(北京)科技有限公司 一种字符分割方法、装置以及计算机可读存储介质
CN111291794A (zh) * 2020-01-21 2020-06-16 上海眼控科技股份有限公司 字符识别方法、装置、计算机设备和计算机可读存储介质
CN111524045A (zh) * 2020-04-13 2020-08-11 北京猿力教育科技有限公司 一种听写方法及装置
CN112364872A (zh) * 2020-10-29 2021-02-12 四川大学华西医院 一种自动识别安瓿瓶的方法及其装置
CN112733834B (zh) * 2021-03-30 2021-07-06 恒银金融科技股份有限公司 一种基于非矩阵窗方式的字符区域定位方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1482571A (zh) * 2003-04-11 2004-03-17 清华大学 基于单个字符的统计笔迹鉴别和验证方法
CN101436249A (zh) * 2008-12-15 2009-05-20 广东国笔科技股份有限公司 一种获得字符匹配模版的方法及系统
CN102156868A (zh) * 2011-03-31 2011-08-17 汉王科技股份有限公司 图像二值化方法和装置
CN102831416A (zh) * 2012-08-15 2012-12-19 广州广电运通金融电子股份有限公司 一种字符识别方法及相关装置

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102136064A (zh) * 2011-03-24 2011-07-27 成都四方信息技术有限公司 一种图像文字识别系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1482571A (zh) * 2003-04-11 2004-03-17 清华大学 基于单个字符的统计笔迹鉴别和验证方法
CN101436249A (zh) * 2008-12-15 2009-05-20 广东国笔科技股份有限公司 一种获得字符匹配模版的方法及系统
CN102156868A (zh) * 2011-03-31 2011-08-17 汉王科技股份有限公司 图像二值化方法和装置
CN102831416A (zh) * 2012-08-15 2012-12-19 广州广电运通金融电子股份有限公司 一种字符识别方法及相关装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115601376A (zh) * 2022-09-22 2023-01-13 北京华益精点生物技术有限公司（Cn） 图像裁切方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN102831416A (zh) 2012-12-19

Similar Documents

Publication Publication Date Title
WO2014026483A1 (fr) Procédé d'identification de caractères et dispositif approprié
CN110334706B (zh) 一种图像目标识别方法及装置
US9349062B2 (en) Character recognition method and device
JP5837205B2 (ja) 画像領域を使用するテキスト検出
WO2018103608A1 (fr) Procédé de détection de texte, dispositif et support d'enregistrement
JP4410732B2 (ja) 顔画像検出装置、顔画像検出方法および顔画像検出プログラム
EP2783328B1 (fr) Détection de texte à l'aide de composants connectés multicouche comportant des histogrammes
US20140029853A1 (en) Form recognition method and device
JP5907593B2 (ja) 画像認識装置、画像認識方法及びプログラム
CN111353961B (zh) 一种文档曲面校正方法及装置
CN111091124B (zh) 一种书脊文字识别方法
CN110348289B (zh) 一种基于二值图的手指静脉识别方法
CN108154132A (zh) 一种身份证文字提取方法、系统及设备和存储介质
CN110378351B (zh) 印章鉴别方法及装置
US20210124918A1 (en) Method and device for acquiring slant value of slant image, terminal and storage medium
Mammeri et al. Road-sign text recognition architecture for intelligent transportation systems
CN110765903A (zh) 行人重识别方法、装置及存储介质
Liang et al. A new wavelet-Laplacian method for arbitrarily-oriented character segmentation in video text lines
CN106446921A (zh) 一种高压输电线障碍物识别方法及装置
US9002115B2 (en) Dictionary data registration apparatus for image recognition, method therefor, and program
Chen et al. Efficient text localization in born-digital images by local contrast-based segmentation
CN116434071B (zh) 一种规整化建筑掩膜的确定方法、确定装置、设备及介质
Gui et al. A fast caption detection method for low quality video images
EP2866171A2 (fr) Procédé et dispositif de détection d'objet
JP2006285959A (ja) 顔判別装置の学習方法、顔判別方法および装置並びにプログラム

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: 13829373

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: 13829373

Country of ref document: EP

Kind code of ref document: A1