WO2019041424A1 - 验证码识别方法、装置、计算机设备及计算机存储介质 - Google Patents

验证码识别方法、装置、计算机设备及计算机存储介质 Download PDF

Info

Publication number
WO2019041424A1
WO2019041424A1 PCT/CN2017/104291 CN2017104291W WO2019041424A1 WO 2019041424 A1 WO2019041424 A1 WO 2019041424A1 CN 2017104291 W CN2017104291 W CN 2017104291W WO 2019041424 A1 WO2019041424 A1 WO 2019041424A1
Authority
WO
WIPO (PCT)
Prior art keywords
character
verification code
code image
boundary
recognition
Prior art date
Application number
PCT/CN2017/104291
Other languages
English (en)
French (fr)
Inventor
周晶
Original Assignee
上海壹账通金融科技有限公司
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 上海壹账通金融科技有限公司 filed Critical 上海壹账通金融科技有限公司
Publication of WO2019041424A1 publication Critical patent/WO2019041424A1/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
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds

Definitions

  • the present invention relates to the field of computer technologies, and in particular, to a verification code identification method, apparatus, computer device, and computer storage medium.
  • the verification code is a public automatic program that effectively distinguishes the user from the computer.
  • the verification code can effectively prevent others from constantly logging in to the website with a specific program, thereby cracking the malicious behavior of the user's account and password.
  • the current verification code includes character images and voices. Identification, video verification code, etc., because the character image is easy to transmit, it is widely used.
  • the website needs to verify whether the verification code character input by the user is correct, firstly, the verification code is automatically recognized, and then The identified verification code is compared with the verification code character input by the user, thereby ensuring the normal login of the user.
  • each character in the verification code is pre-segmented, and some connection lines and characters are stuck between the characters of the verification code. At this time, there is serious noise interference. The result is that the segmentation is inaccurate, and the operation is difficult at the time of segmentation, resulting in a decrease in the accuracy of the identification verification code.
  • a verification code identification method for solving one or more problems involved in the background art.
  • a verification code identification method includes:
  • the standard character corresponding to the maximum probability is output as the recognition result of the current character.
  • a verification code identifying device comprising:
  • a start boundary obtaining module configured to acquire a starting boundary of a current character in the verification code image, and generate an identification window according to the starting boundary, and calculate a probability that the current character in the identification window is a standard character
  • Identifying a window adjustment module configured to fix a height of the recognition window, and increase a width of the recognition window according to a preset step size, and calculate a probability that a character in the recognition window after the width is increased as a standard character until the recognition
  • the aspect ratio of the window is less than or equal to the first threshold
  • a selection module configured to select a maximum probability among the calculated probabilities, and obtain a standard character corresponding to the maximum probability
  • an output module configured to output a standard character corresponding to the maximum probability as a recognition result of the current character.
  • a computer device comprising a memory, a processor, and computer readable instructions stored on the memory and operative on the processor, the processor executing the instructions to:
  • the standard character corresponding to the maximum probability is output as the recognition result of the current character.
  • a computer storage medium having stored thereon computer readable instructions that, when executed by a processor, implement the following steps:
  • the standard character corresponding to the maximum probability is output as the recognition result of the current character.
  • FIG. 1 is a schematic diagram of an application scenario of a verification code identification method in an embodiment
  • FIG. 2 is a flowchart of a verification code identification method in an embodiment
  • FIG. 3 is a schematic diagram of a first identification window of a next character in an embodiment
  • FIG. 4 is a schematic diagram of a second identification window of a next character in an embodiment
  • FIG. 5 is a schematic diagram of a third identification window of a next character in an embodiment
  • FIG. 6 is a schematic diagram of a fourth identification window of a next character in an embodiment
  • Figure 7 is a flow chart showing the steps of generating a starting boundary in an embodiment
  • Figure 8 is a flow chart showing the steps of character recognition in an embodiment
  • FIG. 9 is a flowchart of a verification code image adjustment step in an embodiment
  • Figure 10 is a flow chart showing the steps of character preprocessing in an embodiment
  • FIG. 11 is a schematic structural diagram of a verification code identifying apparatus in an embodiment
  • FIG. 12 is a schematic structural diagram of a computer device in an embodiment.
  • FIG. 1 provides an application scenario diagram of a verification code identification method in an embodiment, including a verification code identification device and a website server, where the verification code identification device and the website server can communicate, and the verification code identification device can be a conventional server.
  • the computer device, etc., the verification code identification device runs a verification code identification program thereon, and the website server has a corresponding website, and the verification code identification device verifies that the verification code input by the user is correct.
  • Website sends an access request to ensure that the user can access the website set up on the website server.
  • the verification code identification device stores a corresponding verification code image.
  • the verification code identification device After the user inputs the verification code, the verification code identification device recognizes the characters in the verification code image, and then compares the recognized characters with the verification code characters input by the user. If the recognized character is successfully compared with the character input by the user, the verification is passed, and then the access request is sent to the website server.
  • FIG. 2 a flowchart of a verification code identification method is provided. This embodiment is applied to the verification code identification device in FIG. 1 to illustrate that the verification code identification device runs on the verification code identification device.
  • the network verification code recognition program identifies the verification code. The method comprises the following steps:
  • S202 Acquire a starting boundary of a current character in the verification code image, and generate an identification window according to the starting boundary, and calculate a probability that the current character in the recognition window is a standard character.
  • the identification window refers to an identification frame set for each character when the characters in the verification code image are recognized, and the identification frame may be a rectangular identification frame, and the first identification edge of the rectangular identification frame is fixed.
  • the second recognition edge can be moved, and the second identification edge and the first recognition edge can be oppositely disposed.
  • the starting boundary refers to the starting position when the character in the verification code image is recognized, and the starting boundary is the first identifying edge described above.
  • Standard characters refer to data that can represent fixed information. Standard characters can be letters, numbers, or symbols. For example, standard characters can be any of 26 English letters, and can be any of numbers 0-9. Is a symbol, such as a period, comma, or exclamation point.
  • the verification code identification device acquires the verification code image, and further obtains a starting position when the current character is recognized in the verification code image, and the starting position is used as a starting boundary for identifying the current character, and is generated according to the starting boundary.
  • the recognition window identifies the position of the first recognition edge of the window as the position of the start boundary, and then uses the trained model to calculate the probability that the current character located in the recognition window is a standard character.
  • the verification code identifying device acquires a starting boundary when the current character in the verification code image is obtained according to the verification code image, generates an identification window according to the starting boundary, and calculates a current character located in the identification window by using the trained model.
  • the probability of a standard character such as the probability that the current character is the standard character a is 30%, and the probability that the current character is the standard character b is 60%.
  • the probability that the current character is the standard character 3 is 89%, and the probability that the current character is the standard character D is 92% or the like.
  • S204 Fix the height of the recognition window, increase the width of the recognition window according to the preset step size, and calculate the probability that the character in the recognition window after the width increase is a standard character until the aspect ratio of the recognition window is less than or equal to the first threshold.
  • the preset step size refers to the amount of change of the width of the recognition window preset, that is, the movement distance preset when the second recognition edge of the recognition window moves, and the width of the recognition window is increased by the movement of the second recognition edge.
  • the preset step size can be set very small, ensuring that the recognition window moves continuously, and the current character located in the recognition window can be accurately identified.
  • the preset step size can be set to 0.01 cm, 0.02 cm, 0.05 cm, 0.07 cm or 0.1 cm. Wait.
  • the first threshold refers to a preset value of the aspect ratio, and the value of the preset aspect ratio can make the recognition window have a reasonable size, ensuring that at most only one current character is located inside the identification window, and the first threshold can be verified according to the verification.
  • the width of each character in the code is preset, for example, the first threshold may be set to 0.75, 0.8, 0.9, 1, 1.2, or the like.
  • the initially set recognition window is a zeroth recognition window, and the height of the zeroth recognition window is fixed, and the width of the recognition window is changed by moving the second recognition edge according to the preset step size, and the width of the recognition window is changed for the first time.
  • the first recognition window is obtained, and the aspect ratio of the recognition window is greater than the first threshold, the probability that the character located in the first recognition window is each standard character is calculated; and then the second step is continued according to the preset step size.
  • Recognizing the edge obtaining a second recognition window, and at this time, the aspect ratio of the recognition window is still greater than the first threshold, calculating the probability that the character located in the second recognition window is each standard character; continuing to move the second according to the preset step size Recognizing the edge until the aspect ratio of the recognition window is less than or equal to the first threshold, and each time the second recognition edge is moved according to the preset step size to obtain a new recognition window, respectively calculating the current character in the new recognition window as a standard character The probability.
  • the height of the recognition window is fixed, the height of the recognition window is 1 cm, and the initial width of the recognition window is 0.1 cm, and the second recognition edge of the movement recognition window changes the width of the recognition window according to the preset step size of 0.05 cm.
  • a first identification window wherein the height of the first recognition window is 1 cm and the width is 0.15 cm, and the aspect ratio of the first recognition window is greater than the first threshold of 0.8, and the calculation is located in the first recognition window.
  • the character is the probability of each standard character, that is, the character located in the first recognition window can be calculated as each of the 26 uppercase letters, each of the 26 lowercase letters, and each of the 10 numbers.
  • S206 Select a maximum probability among the calculated probabilities, and obtain a standard character corresponding to the maximum probability.
  • the verification code identification device calculates a probability that the current character in the recognition window after the width is increased according to the preset step size is a standard character, and selects a maximum probability among all the calculated probabilities, and the standard character corresponding to the maximum probability This is the recognition result of the current character.
  • the height of the recognition window is fixed, the height of the recognition window is 1 cm, the initial width of the recognition window is 0.1 cm, and the second recognition edge of the movement recognition window changes the width of the recognition window each time according to a preset step size of 0.05 cm.
  • a new recognition window is obtained, and each time a new recognition window is obtained, the probability that the current character in the new recognition window is a standard character is calculated separately, that is, each time a new recognition window is obtained, it can be calculated to be located in the new recognition window.
  • the character is the probability of each of the 26 uppercase letters, each of the 26 lowercase letters, and each of the 10 digits, such as a maximum probability of 98%, in which case the maximum probability corresponds to uppercase.
  • the letter D, then the capital letter D is the result of identifying the current character in the window.
  • the verification code identifying device selects the maximum probability among all the calculated probabilities
  • the standard character corresponding to the maximum probability is the recognition result of the current character, and the recognition result is output.
  • the maximum probability is 98%.
  • the maximum probability corresponds to the uppercase letter D
  • the uppercase letter D is the result of identifying the current character of the window
  • the uppercase letter D is output as the recognition window.
  • the current character located in the identification window is identified by using the trained recognition model.
  • the recognition model obtained through the BP (Back Propagation) neural network algorithm can be used to perform the current character located in the recognition window.
  • Identification the recognition model obtained by the naive Bayes algorithm can also be used to identify the current character located in the recognition window, and the model can be trained by using the RMB (Restrited Boltzmann Machine) model. Identify the current character located within the recognition window.
  • the recognition window is set according to the starting boundary, the height of the recognition window is fixed, the width of the recognition window is increased according to the preset step size, and the width is increased every time. Identifying the probability that the character in the window is a standard character, selecting the maximum probability in the probability that the character in the calculated recognition window is a standard character, and the standard character corresponding to the maximum probability as the recognition result of the current character only needs to be identified according to the initial boundary setting.
  • the window fixedly identifies the height of the window, increases the width of the recognition window according to the preset step size, calculates the probability that the character in the recognition window of each increase width is a standard character, and selects the probability that the character in the calculated recognition window is a standard character.
  • the maximum probability, the standard character corresponding to the maximum probability is used as the recognition result of the current character, and the calculation does not need to pre-segment the character, thereby avoiding the operation of the character segmentation operation with difficulty in operation, and ensuring the recognition accuracy of the current character.
  • a character boundary calculation step is provided, which may be performed after step S208 of the embodiment shown in FIG. 2, that is, the standard character corresponding to the maximum probability is output as the recognition result of the current character. After the step, it may also include:
  • S302 Calculate a starting boundary of the next character according to the width of the recognition window corresponding to the maximum probability and the starting boundary of the current character, and identify the next character in the verification code image according to the starting boundary of the next character until the verification code All character recognition in the image is complete.
  • the verification code identification device selects the maximum probability in the probability that the current character is a standard character in the calculated recognition window according to the recognition window, the width of the recognition window corresponding to the maximum probability is obtained, and the calculation is performed according to the start boundary and the width of the recognition window.
  • the starting boundary of the next character according to the starting boundary of the next character, the next character is provided with an identification window of the next character, and the recognition window of the next character is also provided with a fixed first identifying edge and movable Second recognition edge, next character recognition window
  • the position set by the first recognition edge is the starting boundary of the next character, and the probability that the character in the recognition window of the next character is a standard character is calculated, and the recognition window of the next character is added according to the preset step size.
  • the recognition window of the new next character is obtained, and the aspect ratio of the recognition window of the next character is greater than the threshold of the preset aspect ratio, and the probability that the character in the recognition window of the next character is a standard character is calculated.
  • the width of the recognition window of the next character is less than or equal to the threshold of the preset aspect ratio, and each time a new recognition window of the next character is obtained , calculating the probability that the character in the recognition window of the next character is a standard character, selecting the maximum probability, the standard character corresponding to the maximum probability is the next character, and obtaining the width of the recognition window of the next character and the next
  • the starting boundary of a character can be used to calculate the starting boundary of the next character. According to the starting boundary of the next character, an identification window with the next character is set, and then the recognition is performed. A character until all the characters are on the image verification code recognition is completed.
  • a schematic diagram of a first recognition window of a next character, a schematic diagram of a second recognition window of a next character, a schematic diagram of a third recognition window of the next character, and a character of the next character are respectively provided.
  • a schematic diagram of the fourth recognition window The width of the recognition window corresponding to the maximum probability may be calculated by the sum of the initial width of the recognition window and the width increased by a predetermined step size. In this case, the characters in the recognition window are increased by a predetermined number of steps according to the preset step size. The standard character has the highest probability. For example, the initial width of the recognition window is 0.1 cm, and it is moved ten times according to the preset step size of 0.05 cm.
  • the probability that the character in the recognition window is the standard character is the largest, and the width of the recognition window is increased.
  • the width of the recognition window is 0.1 cm and the width of 0.5 cm is 0.6 cm, and the width of the current character recognition window is 0.6 cm; then, the starting boundary of the current character and the distance of the width of the recognition window can be utilized.
  • the position of the starting boundary of the next character is obtained. For example, the position where the starting boundary of the current character is moved by 0.6 cm is the starting boundary of the recognition window of the next character.
  • the starting boundary a shown in FIG. 3 to FIG. 6 is the starting boundary a of the next character.
  • the next character is set to the next character, and the next character is displayed.
  • the identification window can be represented by S
  • the first recognition edge of the recognition window S of the next character is the start boundary a
  • the first recognition edge is fixed
  • the second recognition edge of the recognition window S is b
  • the preset step size is moved, thereby increasing the width of the recognition window S.
  • the preset value of the aspect ratio of the recognition window set with the next character is 1.3.
  • the first recognition window of the next character is set for the next character.
  • the aspect ratio of the first recognition window is 4, which is greater than a preset value of the aspect ratio of the recognition window, and the probability that the next character located in the first recognition window is a standard character is calculated, for example, the next character can be calculated.
  • the probability of A is 1%, the probability of a is 1.2%, the probability of 1 is 70%, the probability of 1 is 75%, and so on.
  • Moving the second recognition edge b of the first recognition window of the next character, obtaining the second recognition window of the next character is S1, as shown in FIG. 4, the first recognition edge of the second recognition window is still the starting boundary a, The second recognition edge of the second recognition window of the next character is b1, and the aspect ratio of the second recognition window is 2, which is greater than a preset value of the aspect ratio of the recognition window, and the next calculation is located in the second recognition window.
  • the probability that a character is a standard character, such as the probability that the next character is A is calculated to be 0.5%, the probability of B is 1%, the probability of being L is 60%, and the like.
  • the first recognition edge of the third recognition window is still the starting boundary a.
  • the second recognition edge of the third recognition window is b2, and the aspect ratio of the third recognition window is 1.5, which is greater than a preset value of the aspect ratio of the recognition window, and the next character located in the third recognition window is calculated as a standard.
  • the probability of a character such as the probability that the next character is C is calculated to be 0.7%, the probability of being d is 1%, the probability of being L is 68%, and so on.
  • the first recognition edge of the fourth recognition window is still the starting boundary a.
  • the second recognition edge of the fourth recognition window is b3, and the aspect ratio of the third recognition window is 1.25, which is smaller than the preset value of the aspect ratio of the recognition window, and the next character located in the fourth recognition window is calculated as a standard.
  • the probability of a character such as the probability that the next character is U is 97%, the probability of F is 1%, the probability of L is 20%, etc.; the maximum probability in the recognition window is 97%. If the corresponding letter is U, the next character is U.
  • the recognition window corresponding to the maximum probability is the fourth recognition window, and the width of the fourth recognition window is 0.8 cm, and according to the starting boundary a of the character U and The width of the fourth recognition window can be used to obtain the starting boundary of the next character.
  • the starting boundary of the next character can be In the second recognition window b3 of the fourth recognition window, according to the start boundary, an identification window of the next character is set, and then the next character is recognized until all the characters on the verification code are recognized.
  • the preset value of the aspect ratio can also be set to 0.6, 0.8, 1.5, etc.; the probability of calculating the characters in the recognition window as standard characters can be respectively calculated as each of the 26 uppercase letters. Probability, and calculate the probability of each of the 26 lowercase letters separately, and calculate the probability of each number in the numbers 0-9 respectively; for the recognition window, the preset step size can be very small, and thus several recognition windows can be obtained. In the present embodiment, only four identification windows are illustrated, and those skilled in the art should understand that the width of the recognition window can be changed very small, and the setting of the recognition window is not limited thereto.
  • the starting boundary of the next character is calculated according to the width of the recognition window corresponding to the maximum probability and the starting boundary of the current character, and the characters in the verification code image are continuously recognized, and all characters are automatically separated without being separated.
  • the recognition of the next character is realized, and the situation in which the segmentation is inaccurate due to the phenomenon of adhesion between characters and the verification code is inaccurate is avoided.
  • a flowchart of a start boundary generation step may be performed, which may be performed before step S202 in the embodiment shown in FIG. 2, where step S202 is to obtain the current character in the verification code image.
  • the start boundary generation step may include:
  • S702 Identify an edge pixel of the verification code image, and select a vertex pixel of the verification code image according to the edge pixel.
  • the edge pixel point refers to a pixel point of the boundary of the verification code image.
  • the edge pixel points may be connected according to the shape of the verification code image to form a corresponding shape contour.
  • a rectangular contour may be formed, and the square may be formed.
  • the edge pixel points of the identification verification code image may be identified according to the coordinate order, for example, the coordinates of the initial recognition pixel are specified, and the initial recognition pixel may be one of the pixels on the lateral side length of the verification code image, and the abscissa of the pixel is guaranteed.
  • edge pixels identifying the verification code image may also be identified by an edge recognition algorithm, such as a drip algorithm, a differential method, or an optimal operator method, for example,
  • the starting point of the identification can be specified, and the identified path can be specified, and the identified path is identified from the identified starting point until all edge pixels of the verification code image are identified.
  • the vertex pixel refers to a pixel of an intersection of each edge of the edge pixel which can constitute a contour of the shape of the verification code image when the edge pixels of the verification code image are sequentially connected, for example, when the verification code image is a rectangle, the vertex pixel can be Is the pixel of the four vertices of the captcha image. Selecting a vertex pixel of the verification code image according to the edge pixel of the recognized verification code image. Specifically, the vertex pixel of the verification code image may be selected according to the coordinates of the edge pixel, for example, when the edge pixel is The abscissa of the position is unchanged, and the ordinate is changed in order.
  • the current edge pixel is one of the vertex coordinates.
  • the abscissa changes in order, and when the abscissa of the next edge pixel position is the same as the abscissa of the current edge pixel position, and the ordinate changes, the current edge pixel
  • the point is one of the vertex coordinates. It is also possible to directly select vertex pixels in the verification code image in the edge pixel according to the edge pixel of the verification code image detected by the edge recognition algorithm.
  • S704 Generate a first boundary of the verification code image according to the vertex pixel, and use the first boundary as a starting boundary of the first character in the verification code image.
  • the vertex pixel of the verification code image when the vertex pixel of the verification code image is identified, one of the vertex pixels is selected as the first vertex pixel, and each pixel adjacent to the first vertex pixel is sequentially connected to form a first boundary of the verification code image.
  • the characters of the verification code image are horizontally arranged, when the vertex pixels of the verification code image are recognized, one of the vertex pixels is selected as the first vertex pixel, and in the vertical direction, the first Each pixel adjacent to the vertex pixel is sequentially connected to form a first boundary of the verification code image; or when the characters of the verification code image are vertically arranged, when the vertex pixel of the verification code image is recognized, One vertex pixel is used as the first vertex pixel, and each pixel point adjacent to the first vertex pixel is sequentially connected in the lateral direction to form an inspection.
  • the first boundary of the code image For example, when the verification code image is a rectangular image and the characters in the verification code image are horizontally arranged, when the vertex pixel of the verification code image is recognized, the vertex pixel point in the upper left corner is selected as the first vertex pixel, according to the vertical In the straight direction, each pixel adjacent to the first vertex pixel is sequentially connected to form a first boundary of the verification code image; or, when the verification code image is a rectangular image and the characters in the verification code image are vertically arranged When the vertex pixel of the verification code image is recognized, the vertex pixel of the upper left corner is selected as the first vertex pixel, and each pixel adjacent to the first vertex pixel is sequentially connected according to the lateral direction to form a verification code image.
  • the first boundary it should be noted that, when the verification code is recognized, the first character from the left side of the verification code image may be identified. When the first boundary of the verification code is generated according to the vertex pixel, the upper left side of the verification code image may be selected.
  • the pixel of the corner is used as the vertex pixel to generate the first boundary, and the pixel of the lower left corner may be selected as the vertex pixel to generate the first boundary; or the first character from the right side of the verification code image may be used for recognition, according to the vertex pixel
  • the pixel in the upper right corner of the verification code image may be selected as the vertex pixel to generate the first boundary, or the pixel in the lower right corner of the verification code image may be selected as the vertex pixel to generate the first boundary.
  • the first boundary of the generated verification code image is used as the starting boundary of the first character in the verification code image, and an identification window is set according to the starting boundary, thereby identifying the first character located in the identification window.
  • the verification code image may be a rectangular image, or may be an image of another shape, such as a parallelogram, a hexagon, etc., for example, may be a parallelogram verification code graphic, when the shape of the verification code image is a parallelogram
  • the vertex pixel of the verification code image is selected according to the edge pixel, and each pixel adjacent to the vertex pixel is connected according to the vertex pixel in the order, if formed after the connection If the height is smaller than the height of the preset recognition window, the pixel adjacent to the vertex pixel is selected, and the pixel adjacent to the vertex pixel is used as a new vertex pixel, which will be adjacent to the new vertex pixel.
  • the pixels are connected in order until the height of the connected pixels is equal
  • the edge pixel of the verification code image is identified, the vertex pixel of the verification code image is selected according to the edge pixel, and the first boundary of the verification code image is generated according to the vertex pixel,
  • the first boundary is used as a starting boundary of the first character in the verification code image, and an identification window is set according to the starting boundary, the characters in the verification code image are identified, the recognition boundary is selected accurately, and the set recognition window is accurate. Improve the accuracy of verification code identification.
  • a flowchart of a character recognition step is provided.
  • the step may be performed after step S702 of the embodiment shown in FIG. 7.
  • Step S702 is to identify edge pixels of the verification code image.
  • Point, after the step of selecting the vertex pixel of the verification code image according to the edge pixel, the character recognition step may include:
  • S802 Generate a second boundary of the verification code image according to the vertex pixel.
  • the second boundary refers to a termination boundary of all character recognition completions in the verification code image, that is, the boundary of the recognition window of all characters does not exceed the termination boundary, that is, does not exceed the second boundary.
  • the vertex pixel of the verification code image is identified, one of the vertex pixels is selected as the second vertex pixel, and each pixel adjacent to the second vertex pixel is sequentially connected to form a second boundary of the verification code image. It may be that when the characters of the verification code image are arranged horizontally, when the vertex pixels of the verification code image are recognized, one of the vertex pixels is selected as the second vertex pixel, and adjacent to the second vertex pixel in the vertical direction.
  • Each pixel is sequentially connected to form a second boundary of the verification code image; or, when the characters of the verification code image are vertically arranged, when a vertex pixel of the verification code image is recognized, one of the vertex pixels is selected As the second vertex pixel, each pixel point adjacent to the second vertex pixel is sequentially connected in the lateral direction to form a second boundary of the verification code image.
  • the vertex pixel point in the upper right corner is selected as the second vertex pixel, according to the vertical In the straight direction, each pixel adjacent to the second vertex pixel is sequentially connected to form a second boundary of the verification code image; or, when the verification code image is a rectangular image and the characters in the verification code image are vertically arranged
  • the vertex pixel in the upper right corner is selected as the second vertex pixel, and each pixel adjacent to the second vertex pixel is sequentially connected according to the lateral direction to form a verification code image.
  • the second boundary when the vertex pixel of the verification code image is recognized, the vertex pixel point in the upper right corner is selected as the second vertex pixel, according to the vertical In the straight direction, each pixel adjacent to the second vertex pixel is sequentially connected to form a second boundary of the verification code image; or, when the verification code image is a rectangular image and the characters in the verification code image are vertically arranged
  • the vertex pixel in the upper right corner is selected as the second vertex
  • the verification code when the verification code is recognized, the first character from the left in the verification code image can be recognized, and the last character from the left is generated, and the second boundary of the verification code is generated according to the vertex pixel.
  • the pixel in the upper right corner of the verification code image is selected as the vertex pixel to generate the second boundary
  • the pixel in the lower right corner may be selected as the vertex pixel to generate the first boundary; or from the right side in the verification code image
  • a character begins to recognize until the last character from the right
  • the second boundary of the verification code when the second boundary of the verification code is generated according to the vertex pixel, the pixel in the upper left corner of the verification code image may be selected as the vertex pixel to generate the first boundary, or may be selected.
  • the pixel in the lower left corner of the captcha image is used as a vertex pixel to generate a first boundary.
  • S804 Calculate a distance between a starting boundary of the next character and a second boundary of the verification code image.
  • the distance between the start boundary of the next character and the second boundary of the captcha image is calculated based on the position of the start boundary of the next character and the position of the second boundary of the captcha image.
  • the start boundary and the verification code of the next character may be calculated according to the coordinates of the position of the first pixel of the start boundary of the next character and the position of the first pixel of the second boundary of the verification code image.
  • the coordinates of the position of the first pixel of the start boundary of the next character are (1, 1)
  • the coordinates of the position of the first pixel of the second boundary are (1, 5)
  • the unit of the distance is centimeter
  • the next The distance between the starting boundary of the character and the second boundary is 4 cm; or the width between the first boundary and the second boundary is 5 cm, and the width of the recognition window of the first character is correctly recognized as 1 cm, the distance between the starting boundary of the next character and the second boundary is 4 cm.
  • the width of the recognition window of the first character can be correctly recognized as 1 cm
  • the recognition window of the second character can be correctly recognized.
  • the width is 1 cm
  • the distance between the starting boundary of the next character and the second boundary is 3 cm.
  • any pixel coordinate in the starting boundary of the next character may be used, and pixel coordinates corresponding to any pixel in the starting boundary of the next character in the second boundary may be selected. Calculation.
  • the second threshold is a preset value of the distance between the start boundary and the second boundary of the next character.
  • the next character is Start The boundary is adjacent to the second boundary, at which point the last character in the captcha image has been reached.
  • all character recognition in the verification code image is completed.
  • the second threshold is set to 2 cm, when the distance between the start boundary of the next character and the second boundary is less than 2 cm, then the last character in the verification code image has been reached, then the image in the verification code image All character recognition is complete.
  • the second threshold may be 1 cm, 1.2 cm, 2.1 cm or 2.2 cm, or the like.
  • the second boundary is set, and the second threshold is further set, and the distance between the start boundary of the next character and the second boundary is calculated.
  • the distance is less than the second threshold, the character recognition in the verification code image is completed.
  • the second boundary and the starting boundary of the next character it can be determined whether the characters in the verification code image are recognized, the operation is simple, no manual monitoring is needed, labor is saved, and work efficiency is improved.
  • a flowchart of a verification code image adjustment step may be provided, which may be performed after the identification window is generated according to the start boundary in step S202 in the embodiment shown in FIG. 202, that is, after acquiring a starting boundary of a current character in the verification code image, and generating an identification window according to the starting boundary, and calculating, in the probability that the current character in the recognition window is a standard character, generating an identification window according to the starting boundary, including :
  • S902 Calculate an aspect ratio of the verification code image when the height of the verification code image does not match the height of the recognition window.
  • the verification code identifying device acquires a starting boundary of the current character in the verification code image according to the verification code image, and generates an identification window of the character in the verification code image according to the starting boundary, when the height of the verification code image and the height of the identification window are not When matching, the verification code identifying device calculates the aspect ratio of the verification code image. Specifically, after the recognition window is generated, it is detected that the height of the verification code image does not match the height of the recognition window, for example, the height value of the verification code image is larger than the height value of the recognition window, so that the characters located in the verification code image exceed the recognition.
  • the window obtains the aspect ratio of the verification code image according to the ratio of the height to the width of the verification code image.
  • the height of the verification code image does not match the height of the recognition window.
  • the height of the verification code image may be smaller than the height of the recognition window, resulting in the character being too small, so that the recognition window is too small, and the character recognition is inaccurate. .
  • S904 Adjust the height and width of the verification code image according to the aspect ratio of the verification code image.
  • the verification code identification device adjusts the height and width of the verification code image according to the calculated aspect ratio of the verification code image, so that the height of the verification code image matches the width of the identification window, and the height and width of the verification code image are adjusted. , to ensure that the aspect ratio of the verification code image is unchanged, to avoid the deformation of the characters in the verification code image due to the adjustment height and width.
  • the height of the verification code image does not match the height of the recognition window, in order to avoid that the character located in the identification window in the verification code image is larger than the recognition window, the character is missing and the recognition is inaccurate, or the verification code image is located. If the characters in the recognition window are too small, affecting the accuracy of the recognition, the height and width of the verification code image need to be adjusted, and the height and width of the verification code image are adjusted according to the aspect ratio of the verification code image to ensure the verification code image. The characters are not distorted by the resizing, ensuring the accuracy of character recognition in the captcha image.
  • a flowchart of a character pre-processing step may be provided, which may be performed before the calculation of the step S202 in the embodiment of FIG. 2 identifies the probability that the current character in the window is a standard character.
  • Step S202 that is, obtaining a starting boundary of the current character in the verification code image, and generating an identification window according to the starting boundary, and calculating a probability that the current character in the recognition window is a standard character, and the current character in the recognition recognition window is a standard character.
  • the probability of execution before including:
  • S1002 Perform binarization processing on the verification code image.
  • the verification code image Before the characters in the verification code image are identified, the verification code image may be binarized.
  • the verification code image may be binarized by using the Otsu algorithm (Otsu algorithm), and the Bernsen binary value may be used.
  • the algorithm can use the Niblack binarization algorithm.
  • a preset pixel value threshold may be set, the pixel value of each pixel in the verification code image is compared with a preset value, and then a new pixel value is set for each pixel according to the comparison result, and further Obtaining a binarized image of the verification code image, for example, the preset pixel value threshold is 155, comparing the pixel value of each pixel in the verification code image with a preset value, when the pixel of the pixel in the verification code image When the value is higher than 155, the pixel value of the pixel is set to 1. When the pixel value of the pixel in the verification code image is lower than 155, the pixel value of the pixel is set to 0, and then the binarization process is obtained. Captcha image. It should be noted that the corresponding settings can be based on the verification code image. Threshold, the threshold can be set to 65, 80, 90, 165, and so on.
  • S1004 Acquire an edge of each character in the verification code image after binarization processing.
  • the edge of each character is obtained.
  • the edge of each character can be obtained by using the bug method, and the canny edge can also be used.
  • the detection operator obtains the edge of each character, and the edge of each character can also be obtained by using the laplacian operator.
  • the detection starting point in the image can be set in advance, thereby specifying the path of detecting the edge of the character, such as from the white pixel area to the black.
  • the pixel area advances, and the black pixel area represents a closed contour.
  • the verification code identification device smoothes the edge of each character according to the edge of each character obtained.
  • the edge of the character may be smoothed by using an exponential smoothing algorithm, and the Laplace algorithm may be used for the character.
  • the edge is smoothed, and the edge of the character can be smoothed by the neighborhood averaging method. For example, after the verification code recognition device acquires the edge of each character, some sawtooth on the edge of the character is smoothed, and the size is small. The missing is filled.
  • the verification code identification device first performs binarization processing on the verification code image, and then extracts the edge of the character in the verification code image, and processes the edge of the character to avoid the influence of the verification code image of different colors on the recognition result. And processing the edge of the character in the verification code image to prevent the recognition result from being affected by the defect of the character edge itself, the accuracy of the character recognition can be improved, and the efficiency of the character recognition can be improved.
  • steps in the flowcharts of FIGS. 2, 7, 8, 9, and 10 above are displayed once in accordance with the indication of the arrows, these steps are not necessarily performed once in the order indicated by the arrows. Except as explicitly stated herein, the execution of these steps is not strictly limited, and may be performed in other sequences. Moreover, at least some of the steps in FIG. 2, FIG. 7, FIG. 8, FIG. 9, and FIG. 10 may include a plurality of sub-steps or stages, which are not necessarily performed at the same time, but may be Execution at different times, the order of execution is not necessarily This is done sequentially, but may be performed alternately or alternately with at least a portion of the other steps or sub-steps or stages of the other steps.
  • the verification code identifying apparatus 110 includes:
  • the start boundary obtaining module 111 is configured to obtain a start boundary of a current character in the verification code image, and generate an identification window according to the start boundary, and calculate a probability that the current character in the recognition window is a standard character.
  • the recognition window adjustment module 112 is configured to fix the height of the recognition window, increase the width of the recognition window according to the preset step size, and calculate the probability that the character in the recognition window after the width increase is a standard character until the aspect ratio of the recognition window is increased. Less than or equal to the first threshold.
  • the selecting module 113 is configured to select a maximum probability among the calculated probabilities and obtain a standard character corresponding to the maximum probability.
  • the output module 114 is configured to output the standard character corresponding to the maximum probability as the recognition result of the current character.
  • the verification code identifying apparatus 110 may further include:
  • a boundary calculation module configured to calculate a starting boundary of a next character according to a width of the recognition window corresponding to the maximum probability and a starting boundary of the current character, and identify a next character in the verification code image according to a starting boundary of the next character Until all character recognition in the captcha image is completed.
  • the verification code identifying apparatus 110 may further include:
  • the vertex pixel selection module is configured to identify an edge pixel of the verification code image, and select a vertex pixel of the verification code image according to the edge pixel.
  • a first boundary generating module configured to generate a first boundary of the verification code image according to the vertex pixel, and use the first boundary as a starting boundary of the first character in the verification code image.
  • the verification code identifying apparatus may further include:
  • a second boundary generation module configured to generate a second boundary of the verification code image according to the vertex pixel.
  • the distance calculation module is configured to calculate a distance between a start boundary of the next character and a second boundary of the verification code image.
  • the recognition completion recording module is configured to, when the distance is less than the second threshold, complete recognition of all characters in the verification code image.
  • the start boundary acquisition module 111 may include:
  • the aspect ratio calculation unit is configured to calculate an aspect ratio of the verification code image when the height of the verification code image does not match the height of the recognition window.
  • An image adjustment unit configured to adjust a height and a width of the verification code image according to an aspect ratio of the verification code image.
  • the verification code identifying apparatus 110 may further include:
  • the graphic binarization processing module is configured to perform binarization processing on the verification code image.
  • the character edge obtaining module is configured to obtain an edge of each character in the verification code image after the binarization process.
  • a character edge processing module is configured to smooth the edge of each character obtained.
  • each of the above-described verification code identification devices may be implemented in whole or in part by software, hardware, and combinations thereof.
  • Each of the above modules may be embedded in or independent of the processor in the computer device, or may be stored in a memory in the computer device in a software form, so that the processor invokes the operations corresponding to the above modules.
  • the processor can be a central processing unit (CPU), a microprocessor, a microcontroller, or the like.
  • the above verification code identifying means may be implemented in the form of a computer readable instruction which may be run on a verification code identifying device as shown in FIG.
  • the embodiment of the present invention provides a computer device, which includes a series of computer readable instructions stored in a memory, and when the computer readable instructions are executed by the processor, the update of the teller machine control according to various embodiments of the present invention may be implemented.
  • the method in some embodiments, is based on the particular operations implemented by the various portions of the computer readable instructions.
  • 12 is a schematic structural diagram of a computer device for performing verification code identification.
  • the computer device may be the above-mentioned verification code identification device, a conventional server or any other suitable computer device, and the internal structure of the computer device may correspond to 12, wherein the computer device includes a memory connected through a system bus, A processor, an operating system, a database, and a verification code identification program stored on the memory and operable on the processor, wherein the processor is configured to provide computing and control capabilities to support operation of the entire computer device.
  • the memory is used to store data, program code, and the like.
  • At least one computer readable instruction is stored on the memory, the computer readable instructions being executable by a processor to implement the verification code identification method provided in various embodiments of the present application.
  • the memory can include internal memory that provides a cached operating environment for operating systems, databases, and computer executable instructions in the non-volatile storage medium.
  • the processor performs the following steps: obtaining a starting boundary of a current character in the verification code image, and generating an identification window according to the starting boundary, and calculating a probability that the current character in the recognition window is a standard character.
  • the height of the recognition window is fixed, and the width of the recognition window is increased according to the preset step size, and the probability that the character in the recognition window after the width increase is a standard character is calculated until the aspect ratio of the recognition window is less than or equal to the first threshold.
  • the maximum probability among the calculated probabilities is selected and the standard characters corresponding to the maximum probability are obtained.
  • the standard character corresponding to the maximum probability is output as the recognition result of the current character.
  • the processor executes the readable instruction
  • the following steps are further implemented: calculating a starting boundary of the next character according to the width of the recognition window corresponding to the maximum probability and the starting boundary of the current character, and according to the next character
  • the starting boundary identifies the next character in the captcha image until all character recognition in the captcha image is complete.
  • the processor when the processor executes the readable instructions, the following steps are further performed: identifying edge pixel points of the verification code image, and selecting vertex pixel points of the verification code image according to the edge pixel points.
  • a first boundary of the verification code image is generated according to the vertex pixel, and the first boundary is used as a starting boundary of the first character in the verification code image.
  • the processor further implements the step of generating a second boundary of the captcha image from the vertex pixels when the processor executes the readable instructions.
  • the distance between the starting boundary of the next character and the second boundary of the captcha image is calculated. When the distance is less than the second threshold, then all character recognition in the verification code image is completed.
  • the processor further implements the following steps when executing the readable instructions: when verifying When the height of the code image does not match the height of the recognition window, the aspect ratio of the verification code image is calculated. The height and width of the captcha image are adjusted according to the aspect ratio of the captcha image.
  • the processor further implements the step of performing binarization processing on the verification code image when the processor executes the readable instructions. Obtain the edge of each character in the binarized verification code image. Smooth the edges of each character obtained.
  • a computer storage medium having stored thereon computer readable instructions, such as the nonvolatile storage medium shown in FIG. 12, wherein the memory can include a disk, an optical disk, a read only memory.
  • Non-volatile storage media such as Read-Only Memory (ROM).
  • the memory includes a non-volatile storage medium and an internal memory.
  • a non-volatile storage medium of a computer device stores an operating system, a database, and computer executable instructions.
  • the database stores data related to implementing a method for updating the teller machine control provided by the various embodiments described above.
  • the following steps are implemented: acquiring a starting boundary of a current character in the verification code image, and generating an identification window according to the starting boundary, and calculating a probability that the current character in the recognition window is a standard character.
  • the height of the recognition window is fixed, and the width of the recognition window is increased according to the preset step size, and the probability that the character in the recognition window after the width increase is a standard character is calculated until the aspect ratio of the recognition window is less than or equal to the first threshold.
  • the maximum probability among the calculated probabilities is selected and the standard characters corresponding to the maximum probability are obtained.
  • the standard character corresponding to the maximum probability is output as the recognition result of the current character.
  • the following steps may be further implemented: calculating a starting boundary of the next character according to the width of the recognition window corresponding to the maximum probability and the starting boundary of the current character, and according to the next The starting boundary of the character identifies the next character in the captcha image until all character recognition in the captcha image is complete.
  • the following steps may be further performed: identifying edge pixel points of the verification code image, and selecting vertex pixel points of the verification code image according to the edge pixel points. Generating a first boundary of the verification code image according to the vertex pixel, and using the first boundary as the verification code image The starting boundary of the first character.
  • the instructions when executed by the processor, may further implement the step of generating a second boundary of the captcha image from the vertex pixels. The distance between the starting boundary of the next character and the second boundary of the captcha image is calculated. When the distance is less than the second threshold, then all character recognition in the verification code image is completed.
  • the instruction may be further executed by the processor to calculate an aspect ratio of the verification code image when the height of the verification code image does not match the height of the recognition window.
  • the height and width of the captcha image are adjusted according to the aspect ratio of the captcha image.
  • the following step may also be implemented: binarizing the verification code image. Obtain the edge of each character in the binarized verification code image. Smooth the edges of each character obtained.
  • the computer readable storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or the like.

Landscapes

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

Abstract

一种验证码识别方法、装置、计算机设备及计算机存储介质，该方法包括获取验证码图像中当前字符的起始边界，并根据起始边界生成识别窗口，计算识别窗口中的当前字符为标准字符的概率；固定识别窗口的高度，且按照预设步长增加识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至识别窗口的高宽比小于等于第一阈值；选取所计算的概率中的最大概率，并获取与最大概率对应的标准字符；将最大概率对应的标准字符作为当前字符的识别结果输出。

Description

验证码识别方法、装置、计算机设备及计算机存储介质
本申请要求于2017年8月28日提交中国专利局、申请号为201710752752.7、发明名称为“验证码识别方法、装置、计算机设备及存储介质”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
本发明涉及计算机技术领域，特别是涉及一种验证码识别方法、装置、计算机设备及计算机存储介质。
背景技术
验证码是一种有效区分用户还是计算机的公共全自动程序，使用验证码可以有效防止他人对网站用特定程序不断登录，从而破解用户的账户和密码等恶意行为，目前验证码包括字符图像、语音识别、视频验证码等，由于字符图像易传输，使用较为广泛，一般当用户进行登陆时输入验证码之后，网站需要验证用户输入的验证码字符是否正确，则首先需要自动识别出验证码，然后将识别出的验证码与用户输入的验证码字符进行比对，从而保证用户的正常登陆。
一般地，在自动识别验证码的时候，会对验证码中的各字符进行预先分割，有些验证码的字符之间会出现一些连接线和字符的黏连，此时，会有严重的噪声干扰到导致分割不准确，在分割时操作难度大，导致识别验证码的准确性降低。
发明内容
根据本申请的各种实施例，提供一种验证码识别方法、装置、计算机设备及计算机存储介质，解决了背景技术中所涉及的一个或多个问题。
一种验证码识别方法，包括：
获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
选取所计算的概率中的最大概率，并获取与所述最大概率对应的标准字符；及
将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
一种验证码识别装置，所述装置包括：
起始边界获取模块，用于获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
识别窗口调节模块，用于固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
选取模块，用于选取所计算的概率中的最大概率，并获取与所述最大概率对应的的标准字符；
输出模块，用于将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
一种计算机设备，包括存储器、处理器以及存储在存储器上并可在处理器上运行的计算机可读指令，所述处理器执行所述指令时，实现以下步骤：
获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
选取所计算的概率中的最大概率，并获取与所述最大概率对应的标准字 符；
将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
一种计算机存储介质，其上存储有计算机可读指令，该指令被处理器执行时实现以下步骤：
获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
选取所计算的概率中的最大概率，并获取与所述最大概率对应的标准字符；
将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
本发明的一个或多个实施例的细节在下面的附图和描述中提出。本发明的其它特征、目的和优点将从说明书、附图以及权利要求书变得明显。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为一实施例中验证码识别方法应用场景图；
图2为一实施例中验证码识别方法的流程图；
图3为一实施例中下一字符的第一识别窗口的示意图；
图4为一实施例中下一字符的第二识别窗口的示意图；
图5为一实施例中下一字符的第三识别窗口的示意图；
图6为一实施例中下一字符的第四识别窗口的示意图；
图7为一实施例中起始边界生成步骤的流程图；
图8为一实施例中字符识别步骤的流程图；
图9为一实施例中验证码图像调节步骤的流程图；
图10为一实施例中字符预处理步骤的流程图；
图11为一实施例中验证码识别装置的结构示意图；
图12为一实施例中计算机设备的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用于解释本发明，并不用于限定本发明。
在详细说明根据本发明的实施例前，应该注意到的是，所述的实施例主要在于与验证码识别方法、装置、计算机设备及计算机存储介质相关的步骤和装置组件的组合。因此，所述装置组件和方法步骤已经在附图中通过常规符号在适当的位置表示出来了，并且只示出了与理解本发明的实施例有关的细节，以免因对于得益于本发明的本领域普通技术人员而言显而易见的那些细节模糊了本发明的公开内容。
在本文中，诸如左和右，上和下，前和后，第一和第二之类的关系术语仅仅用来区分一个实体或动作与另一个实体或动作，而不一定要求或暗示这种实体或动作之间的任何实际的这种关系或顺序。术语“包括”、“包含”或任何其他变体旨在涵盖非排他性的包含，由此使得包括一系列要素的过程、方法、物品或者设备不仅包含这些要素，而且还包含没有明确列出的其他要素，或者为这种过程、方法、物品或者设备所固有的要素。
请参见图1，图1提供一实施例中验证码识别方法应用场景图，其中包括验证码识别设备和网站服务器，验证码识别设备和网站服务器可以进行通信，验证码识别设备可以是常规服务器、计算机设备等，验证码识别设备其上运行有验证码识别程序，网站服务器上设置有相应的网站，验证码识别设备验证用户输入的验证码是否正确，当用户输入的验证码正确时，则向网站 服务器发送访问请求，确保用户可以访问设置在网站服务器上的网站。验证码识别设备中存储有相应的验证码图像，用户输入验证码后，验证码识别设备识别出验证码图像中的字符，然后将识别出的字符与用户输入的验证码字符进行比对，当识别出的字符与用户输入的字符比对成功，则验证通过，进而向网站服务器发送访问请求。
请参见图2，提供一验证码识别方法的流程图，本实施例以该方法应用到上述图1中的验证码识别设备来举例说明，该验证码识别设备上运行有验证码识别程序，通过该网验证码识别程序来识别验证码。该方法包括如下步骤：
S202：获取验证码图像中当前字符的起始边界，并根据起始边界生成识别窗口，计算识别窗口中的当前字符为标准字符的概率。
具体地，识别窗口是指当对验证码图像中的字符进行识别时，针对每个字符设置的识别框，该识别框可以是矩形识别框，该矩形识别框的第一识别边固定不变，第二识别边可以移动，第二识别边与第一识别边可以相对设置。起始边界是指在对验证码图像中的字符进行识别时，对当前字符识别时的起始位置，该起始边界即上述的第一识别边。标准字符是指可以表征固定信息的数据，标准字符可以是字母、数字或符号等，例如，标准字符可以是26个英文字母中的任意一个，可以是数字0-9中的任意一个，也可以是符号，如句号、逗号或感叹号等。
具体地，验证码识别设备获取到验证码图像，进而获取验证码图像中对当前字符进行识别时的起始位置，该起始位置作为识别当前字符的起始边界，根据该起始边界，生成识别窗口，识别窗口的第一识别边设置的位置即为起始边界的位置，进而采用训练得到的模型计算位于识别窗口内的当前字符是标准字符的概率。例如，验证码识别设备根据验证码图像获取验证码图像中的当前字符的识别时的起始边界，根据起始边界，生成识别窗口，采用训练得到的模型计算位于识别窗口内的当前字符为每个标准字符的概率，如当前字符为标准字符a的概率是30％，当前字符为标准字符b的概率是60％， 当前字符为标准字符3的概率是89％，当前字符为标准字符D的概率是92％等。
S204：固定识别窗口的高度，且按照预设步长增加识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至识别窗口的高宽比小于等于第一阈值。
具体地，预设步长是指识别窗口预先设置的宽度的改变量，也即识别窗口的第二识别边移动时预先设置的移动距离，通过第二识别边的移动，增加识别窗口的宽度，预设步长可以设置的非常小，保证识别窗口连续移动，可以准确识别位于识别窗口内的当前字符，例如，预设步长可以设置为0.01厘米、0.02厘米、0.05厘米、0.07厘米或0.1厘米等。第一阈值是指预先设置的高宽比的值，该预先设置的高宽比的值可以使得识别窗口具有合理的大小，保证最多仅有一个当前字符位于识别窗口内部，第一阈值可以根据验证码中每个字符的宽度进行预先设置，例如，第一阈值可以设置为0.75、0.8、0.9、1、1.2等。
具体地，初始设置的识别窗口为第零识别窗口，该第零识别窗口的高度固定，按照预设步长，通过移动第二识别边改变识别窗口的宽度，当第一次改变识别窗口的宽度得到第一识别窗口时，且此时识别窗口的高宽比大于第一阈值，则计算位于第一识别窗口内的字符为每个标准字符的概率；进而按照预设步长，继续移动第二识别边，得到第二识别窗口，且此时识别窗口的高宽比仍大于第一阈值，计算位于第二识别窗口内的字符为每个标准字符的概率；继续按照预设步长移动第二识别边，直至识别窗口的高宽比小于等于第一阈值，且每次按照预设步长移动第二识别边得到新的识别窗口后，分别计算位于新的识别窗口中的当前字符为标准字符的概率。例如，将识别窗口的高度固定，识别窗口的高度为1厘米，此时识别窗口的初始宽度为0.1厘米，按照预设步长0.05厘米，移动识别窗口的第二识别边改变识别窗口的宽度得到第一识别窗口，此时第一识别窗口的高度为1厘米，宽度为0.15厘米，第一识别窗口的高宽比大于第一阈值0.8，计算位于第一识别窗口内 的字符为每个标准字符的概率，即可以计算位于第一识别窗口内的字符为26个大写字母中的每个字母、26个小写字母中的每个字母以及10个数字中的每个数字的概率；进而继续按照预设步长0.05厘米移动第二识别边，直至识别窗口的高宽比小于等于第一阈值0.8时停止，且每次按照预设步长移动第二识别边得到新的识别窗口后，分别计算位于新的识别窗口中的当前字符为标准字符的概率，即可以计算位于新的识别窗口内的字符为26个大写字母中的每个字母、26个小写字母中的每个字母以及10个数字中的每个数字的概率。
S206：选取所计算的概率中的最大概率，并获取与最大概率对应的标准字符。
具体地，验证码识别设备计算每次按照预设步长增加宽度后的识别窗口中的当前字符为标准字符的概率，选择计算出的所有的概率中的最大概率，该最大概率对应的标准字符即为当前字符的识别结果。例如，将识别窗口的高度固定，识别窗口的高度为1厘米，该识别窗口的初始宽度为0.1厘米，每次按照预设步长0.05厘米，移动识别窗口的第二识别边改变识别窗口的宽度得到新的识别窗口，且每次得到新的识别窗口后，分别计算位于新的识别窗口中的当前字符为标准字符的概率，即每次得到新的识别窗口，可以计算位于新的识别窗口内的字符为26个大写字母中的每个字母、26个小写字母中的每个字母以及10个数字中的每个数字的概率，如最大概率为98％，此时该最大概率对应的为大写字母D，则该大写字母D即为识别窗口内的当前字符的结果。
S208：将最大概率对应的标准字符作为当前字符的识别结果输出。
具体地，当验证码识别设备选择计算出的所有的概率中的最大概率，该最大概率对应的标准字符即为当前字符的识别结果，将该识别结果输出。例如，最大概率为98％，此时该最大概率对应的为大写字母D，则该大写字母D即为识别窗口的当前字符的结果，将该大写字母D作为识别窗口输出。
需要说明的是，本实施例中，对每个识别窗口内的字符进行识别时，可 以采用经过训练得到的识别模型对位于识别窗口内的当前字符进行识别，如可以采用经过BP(Back Propagation，反向传播)神经网络算法经过训练得到的识别模型对位于识别窗口内的当前字符进行识别，也可以采用朴素贝叶斯算法经过训练得到的识别模型对位于识别窗口内的当前字符进行识别，还可以采用RMB(Restrited Boltzmann Machine，受限玻尔兹曼机)模型经过训练得到识别模型对位于识别窗口内的当前字符进行识别。
上述实施例中，只需获取验证码图像中当前字符的起始边界，根据起始边界设置识别窗口，固定识别窗口的高度，按照预设步长增加识别窗口的宽度，计算每次增加宽度的识别窗口中的字符为标准字符的概率，选取计算的识别窗口中的字符为标准字符的概率中的最大概率，将该最大概率对应的标准字符作为当前字符的识别结果只需根据初始边界设置识别窗口，固定识别窗口的高度，按照预设步长增加识别窗口的宽度，计算每次增加宽度的识别窗口中的字符为标准字符的概率，选取计算的识别窗口中的字符为标准字符的概率中的最大概率，将该最大概率对应的标准字符作为当前字符的识别结果，计算无需预先分割字符，避免了操作难度大的字符分割操作，保证当前字符的识别准确性，
在其中一个实施例中，提供一字符边界计算步骤，该步骤可以在图2所示实施例的步骤S208之后执行，步骤S208，即在将最大概率对应的标准字符作为当前字符的识别结果输出的步骤之后，还可以包括：
S302：根据最大概率对应的识别窗口的宽度以及当前字符的起始边界，计算下一字符的起始边界，并根据下一字符的起始边界识别验证码图像中的下一字符，直至验证码图像中的所有字符识别完成。
当验证码识别设备根据识别窗口，选取到计算的识别窗口内当前字符为标准字符的概率中的最大概率，进而获取最大概率对应的识别窗口的宽度，根据起始边界和识别窗口的宽度，计算下一字符的起始边界，根据下一字符的起始边界，对下一字符设置有下一字符的识别窗口，下一字符的识别窗口也设置有固定不变的第一识别边和可移动的第二识别边，下一字符识别窗口 的第一识别边设置的位置即为下一字符的起始边界，计算该下一字符的识别窗口内的字符为标准字符的概率，按照预设的步长，增加下一字符的识别窗口的宽度，得到新的下一字符的识别窗口，且该下一字符的识别窗口的高宽比大于预设的高宽比的阈值，计算该下一字符的识别窗口内的字符为标准字符的概率，进而继续增加下一字符的识别窗口的宽度，直至下一字符的识别窗口的高宽比的阈值小于等于预设的高宽比的阈值，且每次得到新的下一字符的识别窗口时，都计算下一字符的识别窗口内的字符为标准字符的概率，选择最大的概率，最大概率对应的标准字符即为该下一字符，且获取该下一字符的识别窗口的宽度与该下一字符的起始边界，可以计算再下一字符的起始边界，根据再下一字符的起始边界，设置有再下一字符的识别窗口，进而识别再下一字符，直至验证码图像上的所有字符都识别完成。
具体地，可参见图3至图6，分别提供下一字符的第一识别窗口的示意图，下一字符的第二识别窗口的示意图，下一字符的第三识别窗口的示意图和下一字符的第四识别窗口的示意图。最大概率对应的识别窗口的宽度可以由识别窗口的初始宽度与按照预设步长增加若干次的宽度的和来计算，此时按照预设步长增加若干次的宽度的识别窗口内的字符为标准字符的概率最大，例如，识别窗口的初始宽度为0.1厘米，按照预设步长0.05厘米移动了十次，此时识别窗口内的字符为标准字符的概率最大，则该识别窗口的宽度增加了0.5厘米，识别窗口的宽度为0.1厘米与0.5厘米的和，为0.6厘米，则当前字符识别窗口的宽度为0.6厘米；然后，可以利用当前字符的起始边界以及识别窗口的宽度移动的距离得到下一字符的起始边界的位置，例如，当前字符的起始边界移动0.6厘米所在的位置即为下一字符的识别窗口的起始边界。
如图3至图6所示的起始边界a即为下一字符的起始边界a，根据下一字符的起始边界a，对下一字符设置有下一字符的识别窗口，下一字符的识别窗口可以用S表示，下一字符的识别窗口S的第一识别边即为起始边界a，该第一识别边固定不变，识别窗口S的第二识别边为b，第二识别边可以按照 预设步长移动，进而可以增加识别窗口S的宽度。设置有下一字符的识别窗口的高宽比的预设值为1.3，如图3所示，根据下一字符的起始边界a，对下一字符设置有下一字符的第一识别窗口，且该第一识别窗口的高宽比为4，大于识别窗口的高宽比的预设值，计算位于第一识别窗口内的下一字符为标准字符的概率，如可以计算出该下一字符为A的概率是1％，为a的概率是1.2％，为1的概率是70％，为1的概率是75％等。
移动下一字符的第一识别窗口的第二识别边b，得到下一字符的第二识别窗口为S1，如图4所示，第二识别窗口的第一识别边仍为起始边界a，下一字符的第二识别窗口的第二识别边为b1，且第二识别窗口的高宽比为2，大于识别窗口的高宽比的预设值，计算位于第二识别窗口内的下一字符为标准字符的概率，如计算出下一字符为A的概率是0.5％，为B的概率是1％，为L的概率是60％等。
继续移动下一字符的第二识别窗口的第二识别边b，得到下一字符的第三识别窗口为S2，如图5所示，第三识别窗口的第一识别边仍为起始边界a，第三识别窗口的第二识别边为b2，且第三识别窗口的高宽比为1.5，大于识别窗口的高宽比的预设值，计算位于第三识别窗口内的下一字符为标准字符的概率，如计算出下一字符为C的概率是0.7％，为d的概率是1％，为L的概率是68％等。
继续移动下一字符的第三识别窗口的第二识别边b，得到下一字符的第四识别窗口为S3，如图6所示，第四识别窗口的第一识别边仍为起始边界a，第四识别窗口的第二识别边为b3，且第三识别窗口的高宽比为1.25，小于识别窗口的高宽比的预设值，计算位于第四识别窗口内的下一字符为标准字符的概率，如计算出下一字符为U的概率是97％，为F的概率是1％，为L的概率是20％等；选取识别窗口内最大的概率，即为97％，此时对应的字母为U，则该下一字符为U，此时该最大概率对应的识别窗口为第四识别窗口，第四识别窗口的宽度为0.8厘米，且根据该字符U的起始边界a与第四识别窗口的宽度，即可得到再下一字符的起始边界，此时再下一字符的起始边界可 以是第四识别窗口的第二识别边b3，根据该起始边界，设置再下一字符的识别窗口，进而识别再下一字符，直至验证码上的所有字符都识别完毕。
需要说明的是，高宽比的预设值可以还可以设置为0.6、0.8、1.5等；对位于识别窗口内的字符计算为标准字符的概率可以分别计算为26个大写字母中每个字母的概率，并分别计算为26个小写字母中每个字母的概率，并分别计算为数字0-9中每个数字的概率；对于识别窗口，预设步长可以非常小，进而可以得到若干识别窗口，本实施例中仅举出四个识别窗口进行说明，本领域的技术人员应当理解，识别窗口的宽度可以改变的非常小，识别窗口的设置不限于此。
上述实施例中，根据最大概率对应的识别窗口的宽度及当前字符的起始边界，计算下一个字符的起始边界，并继续识别验证码图像中的字符，无需分割出所有字符，即可自动实现下一字符的识别，避免字符之间出现黏连等现象导致的分割不准确从而验证码识别不准确的情形。
在其中一个实施例中，可参见图7，提供一起始边界生成步骤的流程图，该步骤可以在图2所示实施例中步骤S202之前执行，步骤S202，即获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率的步骤之前执行，该起始边界生成步骤可以包括：
S702：识别验证码图像的边缘像素点，根据边缘像素点选取验证码图像的顶点像素点。
具体地，边缘像素点是指验证码图像的边界的像素点，根据验证码图像的形状，边缘像素点相连接可以根据验证码图像的形状构成相应的形状轮廓，例如，可以构成矩形轮廓，正方形轮廓、平行四边形轮廓或五边形轮廓等。具体地，识别验证码图像的边缘像素点可以按照坐标顺序识别，例如，规定起始识别像素的坐标，起始识别像素可以是验证码图像横向边长上的像素之一，保证像素的横坐标不变，顺序改变纵坐标，按照顺序改变纵坐标时，当识别到下一个坐标处无像素点，进而保持识别像素的横坐标不变，改变纵坐 标继续顺序识别，直至识别出验证码图像的所有边缘像素点；识别验证码图像的边缘像素点还可以采用边缘识别算法进行识别，如采用滴水算法、微分法或最优算子法，例如，可以规定识别的起点，进而规定识别的路径，按照识别的路径从识别的起点进行识别，直至验证码图像的所有边缘像素点被识别出来。
顶点像素点是指验证码图像的边缘像素点顺序连接时可以构成验证码图像的形状的轮廓的边缘像素点中的各个边的交点的像素，例如，当验证码图像为矩形时，顶点像素可以是验证码图像的四个顶点的像素。根据识别出的验证码图像的边缘像素点，选取验证码图像的顶点像素点，具体地，可以根据边缘像素点的坐标，选择验证码图像的顶点像素点，例如，可以是，当边缘像素点位置的横坐标不变，纵坐标按顺序改变，当下一个边缘像素点位置的纵坐标与当前边缘像素点位置的纵坐标相同，而横坐标改变，则当前边缘像素点即为顶点坐标之一，也可以是，当边缘像素点位置的纵坐标不变，横坐标按顺序改变，当下一个边缘像素点位置的横坐标与当前边缘像素点位置的横坐标相同，而纵坐标改变，则当前边缘像素点即为顶点坐标之一。还可以根据采用边缘识别算法检测出的验证码图像的边缘像素点，直接选择边缘像素点中的验证码图像中的顶点像素。
S704：根据顶点像素生成验证码图像的第一边界，将第一边界作为验证码图像中的第一个字符的起始边界。
具体地，当识别出验证码图像的顶点像素时，选择其中的一个顶点像素作为第一顶点像素，将与第一顶点像素相邻的每个像素点顺序连接，形成验证码图像的第一边界，具体地，可以是，当验证码图像的字符是横向排列时，当识别出验证码图像的顶点像素时，选择其中的一个顶点像素作为第一顶点像素，按照竖直方向，将与第一顶点像素相邻的每个像素点顺序连接，形成验证码图像的第一边界；也可以是，当验证码图像的字符是竖向排列时，当识别出验证码图像的顶点像素时，选择其中的一个顶点像素作为第一顶点像素，按照横向方向，将与第一顶点像素相邻的每个像素点顺序连接，形成验 证码图像的第一边界。例如，可以是，当验证码图像是矩形图像且验证码图像中的字符是横向排列时，当识别出验证码图像的顶点像素时，选择左上角的顶点像素点作为第一顶点像素，按照竖直方向，将与第一顶点像素相邻的每个像素点顺序连接，形成验证码图像的第一边界；也可以是，当验证码图像是矩形图像且验证码图像中的字符是竖向排列时，当识别出验证码图像的顶点像素时，选择左上角的顶点像素点作为第一顶点像素，按照横向方向，将与第一顶点像素相邻的每个像素点顺序连接，形成验证码图像的第一边界。需要说明的是，在对验证码进行识别时，可以从验证码图像中的从左边起第一个字符开始识别，则根据顶点像素生成验证码的第一边界时，可以选取验证码图像的左上角的像素作为顶点像素从而生成第一边界，也可以选择左下角的像素作为顶点像素从而生成第一边界；也可以从验证码图像中的从右边起第一个字符开始识别，则根据顶点像素生成验证码的第一边界时，可以选取验证码图像的右上角的像素作为顶点像素从而生成第一边界，也可以选取验证码图像的右下角的像素作为顶点像素从而生成第一边界。
将生成的验证码图像的第一边界，作为验证码图像中第一个字符的起始边界，根据起始边界，设置识别窗口，进而对位于识别窗口内的第一个字符进行识别。需要说明的是，验证码图像可以是矩形图像，也可以是其他形状的图像，如平行四边形、六边形等，例如，可以是平行四边形的验证码图形，当验证码图像的形状是平行四边形时，可以先识别验证码图像的边缘像素点，根据边缘像素点选取验证码图像的顶点像素点，根据顶点像素点按顺序连接与顶点像素点相邻的每个像素点时，如果连接后形成的高小于预设的识别窗口的高度，则选取与顶点像素点相邻的像素点，将与顶点像素点相邻的像素点作为新的顶点像素点，将与新的顶点像素点相邻的像素点按顺序连接，直至连接后的像素点形成的高与预设的识别窗口高度相等，将连接后的像素点作为第一边界，将该第一边界作为验证码图像中的第一字符的起始边界。
上述实施例中，识别出验证码图像的边缘像素点，根据边缘像素点选取验证码图像的顶点像素点，根据顶点像素生成验证码图像的第一边界，将该 第一边界作为验证码图像中的第一个字符的起始边界，根据该起始边界设置有识别窗口，对验证码图像中的字符进行识别，选取识别边界准确，进而设置的识别窗口准确，提高对验证码识别的准确性。
在其中一个实施例中，可参见图8，提供一字符识别步骤的流程图，该步骤可以在图7所示实施例的步骤S702后执行，步骤S702，即识别所述验证码图像的边缘像素点，根据所述边缘像素点选取所述验证码图像的顶点像素点的步骤之后执行，字符识别步骤可以包括：
S802：根据顶点像素生成验证码图像的第二边界。
具体地，第二边界是指验证码图像中所有字符识别完成的终止边界，也即所有字符的识别窗口的边界不超过该终止边界，即不超过该第二边界。具体地，当识别出验证码图像的顶点像素时，选择其中一个顶点像素作为第二顶点像素，将与第二顶点像素相邻的每个像素点顺序连接，形成验证码图像的第二边界，可以是，当验证码图像的字符是横向排列时，当识别出验证码图像的顶点像素时，选择其中的一个顶点像素作为第二顶点像素，按照竖直方向，将与第二顶点像素相邻的每个像素点顺序连接，形成验证码图像的第二边界；也可以是，当验证码图像的字符是竖向排列时，当识别出验证码图像的顶点像素时，选择其中的一个顶点像素作为第二顶点像素，按照横向方向，将与第二顶点像素相邻的每个像素点顺序连接，形成验证码图像的第二边界。例如，可以是，当验证码图像是矩形图像且验证码图像中的字符是横向排列时，当识别出验证码图像的顶点像素时，选择右上角的顶点像素点作为第二顶点像素，按照竖直方向，将与第二顶点像素相邻的每个像素点顺序连接，形成验证码图像的第二边界；也可以是，当验证码图像是矩形图像且验证码图像中的字符是竖向排列时，当识别出验证码图像的顶点像素时，选择右上角的顶点像素点作为第二顶点像素，按照横向方向，将与第二顶点像素相邻的每个像素点顺序连接，形成验证码图像的第二边界。需要说明的是，在对验证码进行识别时，可以从验证码图像中的从左边起第一个字符开始识别，直到从左边起的最后一个字符，则根据顶点像素生成验证码的第二边界 时，可以选取验证码图像的右上角的像素作为顶点像素从而生成第二边界，也可以选择右下角的像素作为顶点像素从而生成第一边界；也可以从验证码图像中的从右边起的第一个字符开始识别，直到从右边起的最后一个字符，则根据顶点像素生成验证码的第二边界时，可以选取验证码图像的左上角的像素作为顶点像素从而生成第一边界，也可以选取验证码图像的左下角的像素作为顶点像素从而生成第一边界。
S804：计算下一字符的起始边界与验证码图像的第二边界的距离。
根据下一字符的起始边界的位置与验证码图像的第二边界的位置，计算下一字符的起始边界与验证码图像的第二边界间的距离。具体地，可以是，根据下一字符的起始边界的第一像素的位置的坐标与验证码图像的第二边界的第一像素的位置的坐标，计算下一字符的起始边界与验证码图像的第二边界间的距离；也可以是，根据验证码图像的第一边界与第二边界之间的宽度，每个字符的可正确识别出窗口内的字符的识别窗口的宽度的和，计算两者差值，得到下一字符的起始边界与验证码图像的第二边界的距离。例如，下一字符的起始边界的第一像素的位置的坐标为(1，1)，第二边界的第一像素的位置的坐标为(1，5)，距离的单位为厘米，则下一字符的起始边界与第二边界之间的距离为4厘米；也可以是，第一边界与第二边界之间的宽度为5厘米，可正确识别第一个字符的识别窗口的宽度为1厘米，则下一字符的起始边界与第二边界之间的距离为4厘米，若可正确识别第一个字符的识别窗口的宽度为1厘米，可正确识别第二个字符的识别窗口的宽度为1厘米，则下一字符的起始边界与第二边界之间的距离为3厘米。
需要说明的是，采用坐标计算时，可以采用下一字符的起始边界中的任一像素坐标，选取第二边界中的与下一字符的起始边界中的任一像素对应的像素坐标进行计算。
S806：当距离小于第二阈值时，则验证码图像中的所有字符识别完成。
第二阈值是指下一字符的起始边界与第二边界之间的距离的预设值，当下一字符的起始边界与第二边界之间的距离小于预设值时，下一字符的起始 边界临近第二边界，则此时已到验证码图像中的最后一个字符。具体地，当下一字符的起始边界与第二边界之间的距离小于预设值时，则验证码图像中的所有字符识别完成。例如，设置第二阈值为2厘米，当下一字符的起始边界与第二边界之间的距离小于2厘米时，则此时已到验证码图像中的最后一个字符，则验证码图像中的所有字符识别完成。需要说明的是，第二阈值可以是1厘米、1.2厘米、2.1厘米或2.2厘米等。
上述实施例中，设置有第二边界，还设置有第二阈值，计算下一字符的起始边界与第二边界的距离，当距离小于第二阈值时，则验证码图像中的字符识别完成，根据第二边界与下一字符的起始边界即可确定验证码图像中的字符是否识别完成，操作简单，无需人工过多监控，节省人力，提高工作效率。
在其中一个实施例中，可参见图9，提供一验证码图像调节步骤的流程图，该步骤可以在图2所示实施例中的步骤S202中的根据起始边界生成识别窗口之后执行，步骤202，即在获取验证码图像中当前字符的起始边界，并根据起始边界生成识别窗口，计算识别窗口中的当前字符为标准字符的概率中的根据起始边界生成识别窗口之后执行，包括：
S902：当验证码图像的高度与识别窗口的高度不匹配时，计算验证码图像的高宽比。
验证码识别设备根据验证码图像，获取到验证码图像中当前字符的起始边界，根据该起始边界生成验证码图像中的字符的识别窗口，当验证码图像的高度与识别窗口的高度不匹配时，则验证码识别设备计算验证码图像的高宽比。具体地，当生成识别窗口之后，检测到验证码图像的高度与识别窗口的高度不匹配，例如，验证码图像的高度值比识别窗口的高度值大，从而位于验证码图像中的字符超过识别窗口，则根据验证码图像的高度与宽度的比值，得到验证码图像的高宽比。需要说明的是，验证码图像的高度与识别窗口的高度不匹配也可以是验证码图像的高度比识别窗口的高度小，导致字符过小，从而在识别窗口内过小，出现字符识别不准确。
S904：根据验证码图像的高宽比调节验证码图像的高度和宽度。
具体地，验证码识别设备根据计算出的验证码图像的高宽比，调整验证码图像的高度和宽度，使得验证码图像的高度与识别窗口的宽度匹配，调整验证码图像的高度和宽度时，保证验证码图像的高宽比不变，避免验证码图像中的字符由于调节高度和宽度导致变形。
上述实施例中，当验证码图像的高度与识别窗口的高度不匹配时，为了避免验证码图像中位于识别窗口中的字符比识别窗口大导致字符缺失进而识别不准确，或验证码图像中位于识别窗口内的字符过小，影响识别的准确性，则需要调节验证码图像的高度和宽度，按照验证码图像的高宽比，对验证码图像的高度和宽度进行调节，保证验证码图像中的字符不会因为调节大小而变形，保证验证码图像中字符识别的准确性。
在其中一个实施例中，可参见图10，提供一字符预处理步骤的流程图，该步骤可以在图2所示实施例中步骤S202的计算识别窗口中的当前字符为标准字符的概率之前执行，步骤S202，即在获取验证码图像中当前字符的起始边界，并根据起始边界生成识别窗口，计算识别窗口中的当前字符为标准字符的概率的计算识别窗口中的当前字符为标准字符的概率之前执行，包括：
S1002：对验证码图像进行二值化处理。
在对验证码图像中的字符进行识别之前，可以将验证码图像进行二值化处理，可选地，可以采用Otsu算法(大津算法)对验证码图像进行二值化处理，可以采用Bernsen二值化算法，可以采用Niblack二值化算法等。具体地，可以设置有预设的像素值阈值，将验证码图像中的每一个像素点的像素值与预设值进行比较，进而根据比较结果设置对每一个像素点设置新的像素值，进而得到验证码图像的二值化图像，例如，预设的像素值阈值为155，将验证码图像中的每一个像素点的像素值与预设值进行比较，当验证码图像中的像素的像素值高于155时，则将该像素的像素值设置为1，当验证码图像中的像素的像素值低于155时，则将该像素的像素值设置为0，进而得到二值化处理后的验证码图像。需要说明的是，可以根据验证码图像，相应的设置 阈值，阈值可以设置为65、80、90、165等。
S1004：获取二值化处理后的验证码图像中的每个字符的边缘。
将二值化图像处理后，获取每个字符的边缘，具体地，将验证码图像进行二值化处理后，可选地，可以采用虫随法获取每个字符的边缘，也可以采用canny边缘检测算子获取每个字符的边缘，也可以采用laplacian算子获取每个字符的边缘，例如，可以预先设置图像中的检测起始点，进而规定检测字符边缘的路径，如从白色像素区域向黑色像素区域前进，黑色像素区域表示一个闭合的轮廓，当检测到黑色像素时，则向左继续检测，检测到像素为白色时，则向右继续检测，直到回到最初的检测起始点，则黑色像素即为检测出来的识别码的边界。
S1006：将获取到的每个字符的边缘进行平滑处理。
具体地，验证码识别设备根据获取到的每个字符的边缘，将每个字符的边缘进行平滑处理，如可以采用指数平滑算法对字符的边缘进行平滑处理，可以采用拉普拉斯算法对字符的边缘进行平滑处理，还可以采用邻域平均法对字符的边缘进行平滑处理等，例如，验证码识别设备获取到每个字符的边缘后，对字符的边缘上某些锯齿进行平滑，对细小的缺失进行填补。
上述实施例中，验证码识别设备先将验证码图像进行二值化处理，进而提取验证码图像中字符的边缘，对字符的边缘进行处理，避免了不同颜色的验证码图像对识别结果的影响，且对验证码图像中的字符的边缘进行处理，防止由于字符边缘本身的缺陷影响识别结果，可以提高对字符识别的准确性，并且可以提高字符识别的效率。
虽然上文中图2、图7、图8、图9与图10的流程图中的各个步骤按照箭头的指示一次显示，但是这些步骤并不是必然按照箭头指示的顺序一次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，其可以以其他的顺序执行。而且，图2、图7、图8、图9与图10中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，其执行顺序也不必然 是依次进行，而是可以与其他步骤或者其他步骤的子步骤或者阶段的至少一部分轮流或者交替执行。
在其中一个实施例中，可参见图11，提供一验证码识别装置的结构示意图，验证码识别装置110包括：
起始边界获取模块111，用于获取验证码图像中当前字符的起始边界，并根据起始边界生成识别窗口，计算识别窗口中的当前字符为标准字符的概率。
识别窗口调节模块112，用于固定识别窗口的高度，且按照预设步长增加识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至识别窗口的高宽比小于等于第一阈值。
选取模块113，用于选取所计算的概率中的最大概率，并获取与最大概率对应的的标准字符。
输出模块114，用于将最大概率对应的标准字符作为当前字符的识别结果输出。
在其中一个实施例中，验证码识别装置110还可以包括：
边界计算模块，用于根据最大概率对应的识别窗口的宽度以及当前字符的起始边界，计算下一字符的起始边界，并根据下一字符的起始边界识别验证码图像中的下一字符，直至验证码图像中的所有字符识别完成。
在其中一个实施例中，验证码识别装置110还可以包括：
顶点像素点选取模块，用于识别验证码图像的边缘像素点，根据边缘像素点选取验证码图像的顶点像素点。
第一边界生成模块，用于根据顶点像素生成验证码图像的第一边界，将第一边界作为验证码图像中的第一个字符的起始边界。
在其中一个实施例中，验证码识别装置还可以包括：
第二边界生成模块，用于根据顶点像素生成验证码图像的第二边界。
距离计算模块，用于计算下一字符的起始边界与验证码图像的第二边界的距离。
识别完成记录模块，用于当距离小于第二阈值时，则验证码图像中的所有字符识别完成。
在其中一个实施例中，起始边界获取模块111可以包括：
高宽比计算单元，用于当验证码图像的高度与识别窗口的高度不匹配时，计算验证码图像的高宽比。
图像调节单元，用于根据验证码图像的高宽比调节验证码图像的高度和宽度。
在其中一个实施例中，验证码识别装置110还可以包括：
图形二值化处理模块，用于对验证码图像进行二值化处理。
字符边缘获取模块，用于获取二值化处理后的验证码图像中的每个字符的边缘。
字符边缘处理模块，用于将获取到的每个字符的边缘进行平滑处理。
上述关于验证码识别装置的具体限定可以参见上文中关于验证码识别方法的限定，在此不再赘述。上述验证码识别装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。该处理器可以为中央处理单元(CPU)、微处理器、单片机等。上述验证码识别装置可以实现为一种计算机可读指令的形式，计算机可读指令可在如图1所示的验证码识别设备上运行。
本发明实施例提出了一种计算机设备，该计算机设备包括一系列存储于存储器上的计算机可读指令，当计算机可读指令被处理器执行时，可以实现本发明各个实施例提出的柜员机控件更新方法，在一些实施例中，基于该计算机可读指令各部分所实现的特定的操作。请参见图12，提供一执行验证码识别的计算机设备的结构示意图，该计算机设备可以是上述验证码识别设备、是常规服务器或其他任何合适的计算机设备，该计算机设备的内部结构可对应与图12所示的结构，其中该计算机设备包括通过系统总线连接的存储器、 处理器、操作系统、数据库以及存储在存储器上并可在处理器上运行的验证码识别程序，其中，该处理器用于提供计算和控制能力，支撑整个计算机设备的运行。存储器用于存储数据、程序代码等。
该存储器上存储至少一个计算机可读指令，该计算机可读指令可被处理器执行，以实现本申请各实施例中提供的验证码识别方法。存储器可以包括内存储器，内存储器为非易失性存储介质中的操作系统、数据库和计算机可执行指令提供高速缓存的运行环境。
其中，处理器执行该验计算机可读指令时实现以下步骤：获取验证码图像中当前字符的起始边界，并根据起始边界生成识别窗口，计算识别窗口中的当前字符为标准字符的概率。固定识别窗口的高度，且按照预设步长增加识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至识别窗口的高宽比小于等于第一阈值。选取所计算的概率中的最大概率，并获取与最大概率对应的标准字符。将最大概率对应的标准字符作为当前字符的识别结果输出。
在其中一个实施例中，处理器执行可读指令时还实现以下步骤：根据最大概率对应的识别窗口的宽度以及当前字符的起始边界，计算下一字符的起始边界，并根据下一字符的起始边界识别验证码图像中的下一字符，直至验证码图像中的所有字符识别完成。
在其中一个实施例中，处理器执行可读指令时还实现以下步骤：识别验证码图像的边缘像素点，根据边缘像素点选取验证码图像的顶点像素点。根据顶点像素生成验证码图像的第一边界，将第一边界作为验证码图像中的第一个字符的起始边界。
在其中一个实施例中，处理器执行可读指令时还实现以下步骤：根据顶点像素生成验证码图像的第二边界。计算下一字符的起始边界与验证码图像的第二边界的距离。当距离小于第二阈值时，则验证码图像中的所有字符识别完成。
在其中一个实施例中，处理器执行可读指令时还实现以下步骤：当验证 码图像的高度与识别窗口的高度不匹配时，计算验证码图像的高宽比。根据验证码图像的高宽比调节验证码图像的高度和宽度。
在其中一个实施例中，处理器执行可读指令时还实现以下步骤：对验证码图像进行二值化处理。获取二值化处理后的验证码图像中的每个字符的边缘。将获取到的每个字符的边缘进行平滑处理。
上述关于计算机设备的具体限定可以参见上文中关于验证码识别方法的限定，在此不再赘述。
请继续参阅图12，还提供一种计算机存储介质，其上存储有计算机可读指令，如图12中所示的非易失性存储介质，其中，存储器可包括磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性存储介质等。在一个实施例中，存储器包括非易失性存储介质及内存储器。计算机设备的非易失性存储介质存储有操作系统、数据库和计算机可执行指令。该数据库中存储有用于实现上述各个实施例所提供的一种柜员机控件更新方法相关的数据。其中，该指令被处理器执行时实现以下步骤：获取验证码图像中当前字符的起始边界，并根据起始边界生成识别窗口，计算识别窗口中的当前字符为标准字符的概率。固定识别窗口的高度，且按照预设步长增加识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至识别窗口的高宽比小于等于第一阈值。选取所计算的概率中的最大概率，并获取与最大概率对应的标准字符。将最大概率对应的标准字符作为当前字符的识别结果输出。
在其中一个实施例中，该指令被处理器执行时还可以实现以下步骤：根据最大概率对应的识别窗口的宽度以及当前字符的起始边界，计算下一字符的起始边界，并根据下一字符的起始边界识别验证码图像中的下一字符，直至验证码图像中的所有字符识别完成。
在其中一个实施例中，该指令被处理器执行时还可以实现以下步骤：识别验证码图像的边缘像素点，根据边缘像素点选取验证码图像的顶点像素点。根据顶点像素生成验证码图像的第一边界，将第一边界作为验证码图像中的 第一个字符的起始边界。
在其中一个实施例中，该指令被处理器执行时还可以实现以下步骤：根据顶点像素生成验证码图像的第二边界。计算下一字符的起始边界与验证码图像的第二边界的距离。当距离小于第二阈值时，则验证码图像中的所有字符识别完成。
在其中一个实施例中，该指令被处理器执行时还可以实现以下步骤：当验证码图像的高度与识别窗口的高度不匹配时，计算验证码图像的高宽比。根据验证码图像的高宽比调节验证码图像的高度和宽度。
在其中一个实施例中，该指令被处理器执行时还可以实现以下步骤：对验证码图像进行二值化处理。获取二值化处理后的验证码图像中的每个字符的边缘。将获取到的每个字符的边缘进行平滑处理。
上述关于计算机存储介质的具体限定可以参见上文中关于验证码识别方法的限定，在此不再赘述。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成的指令可存储于一非易失性计算机可读取计算机存储介质中，该指令在执行时，可包括如上述各方法的实施例的流程。其中，计算机可读取的计算机存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本发明的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本发明构思的前提下，还可以做出若干变形和改进，这些都属于本发明的保护范围。因此，本发明专利的保护范围应以所附权利要求为准。

Claims (24)

  1. 一种验证码识别方法，其特征在于，包括：
    获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
    固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
    选取所计算的概率中的最大概率，并获取与所述最大概率对应的标准字符；及
    将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
  2. 根据权利要求1所述的方法，其特征在于，在将所述最大概率对应的标准字符作为所述当前字符的识别结果输出的步骤之后，还包括：
    根据所述最大概率对应的识别窗口的宽度以及所述当前字符的起始边界，计算下一字符的起始边界，并根据所述下一字符的起始边界识别所述验证码图像中的下一字符，直至所述验证码图像中的所有字符识别完成。
  3. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    识别所述验证码图像的边缘像素点，根据所述边缘像素点选取所述验证码图像的顶点像素点；及
    根据所述顶点像素生成所述验证码图像的第一边界，将所述第一边界作为所述验证码图像中的第一个字符的起始边界。
  4. 根据权利要求3所述的方法，其特征在于，所述方法还包括：
    根据所述顶点像素生成所述验证码图像的第二边界；
    计算所述下一字符的起始边界与所述验证码图像的第二边界的距离；及
    当所述距离小于第二阈值时，则所述验证码图像中的所有字符识别完成。
  5. 根据权利要求1所述的方法，其特征在于，所述根据所述起始边界生成识别窗口之后，还包括：
    当所述验证码图像的高度与所述识别窗口的高度不匹配时，计算所述验 证码图像的高宽比；及
    根据所述验证码图像的高宽比调节所述验证码图像的高度和宽度。
  6. 根据权利要求1所述的方法，其特征在于，所述计算所述识别窗口中的当前字符为标准字符的概率之前，还包括：
    对所述验证码图像进行二值化处理；
    获取二值化处理后的验证码图像中的每个字符的边缘；及
    将获取到的每个字符的边缘进行平滑处理。
  7. 一种验证码识别装置，其特征在于，包括：
    起始边界获取模块，用于获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
    识别窗口调节模块，用于固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
    选取模块，用于选取所计算的概率中的最大概率，并获取与所述最大概率对应的的标准字符；及
    输出模块，用于将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
  8. 根据权利要求7所述的装置，其特征在于，所述装置还包括：
    边界计算模块，用于根据所述最大概率对应的识别窗口的宽度以及所述当前字符的起始边界，计算下一字符的起始边界，并根据所述下一字符的起始边界识别所述验证码图像中的下一字符，直至所述验证码图像中的所有字符识别完成。
  9. 根据权利要求7所述的装置，其特征在于，所述装置还包括：
    像素点识别模块，用于识别所述验证码图像的边缘像素点，根据所述边缘像素点选取所述验证码图像的顶点像素点；及
    第一边界生成模块，用于根据所述顶点像素生成所述验证码图像的第一 边界，将所述第一边界作为所述验证码图像中的第一个字符的起始边界。
  10. 根据权利要求9所述的装置，其特征在于，所述装置还包括：
    第二边界生成模块，用于根据所述顶点像素生成所述验证码图像的第二边界；
    距离计算模块，用于计算所述下一字符的起始边界与所述验证码图像的第二边界的距离；及
    识别完成模块，用于当所述距离小于第二阈值时，则所述验证码图像中的所有字符识别完成。
  11. 根据权利要求7所述的装置，其特征在于，所述装置还包括：
    高宽比计算模块，用于当所述验证码图像的高度与所述识别窗口的高度不匹配时，计算所述验证码图像的高宽比；及
    调节模块，用于根据所述验证码图像的高宽比调节所述验证码图像的高度和宽度。
  12. 根据权利要求7所述的装置，其特征在于，所述装置还包括：
    二值化处理模块，用于对所述验证码图像进行二值化处理；
    边缘获取模块，用于获取二值化处理后的验证码图像中的每个字符的边缘；及
    平滑处理模块，用于将获取到的每个字符的边缘进行平滑处理。
  13. 一种计算机设备，包括存储器、处理器以及存储在存储器上并可在处理器上运行的计算机可读指令，其特征在于，所述处理器执行所述指令时实现以下步骤：
    获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
    固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
    选取所计算的概率中的最大概率，并获取与所述最大概率对应的标准字 符；及
    将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
  14. 根据权利要求13所述的计算机设备，其特征在于，所述处理器执行所述将所述最大概率对应的标准字符作为所述当前字符的识别结果输出的步骤之后，还执行以下步骤：
    根据所述最大概率对应的识别窗口的宽度以及所述当前字符的起始边界，计算下一字符的起始边界，并根据所述下一字符的起始边界识别所述验证码图像中的下一字符，直至所述验证码图像中的所有字符识别完成。
  15. 根据权利要求13所述的计算机设备，其特征在于，所述处理器还执行以下步骤：
    识别所述验证码图像的边缘像素点，根据所述边缘像素点选取所述验证码图像的顶点像素点；及
    根据所述顶点像素生成所述验证码图像的第一边界，将所述第一边界作为所述验证码图像中的第一个字符的起始边界。
  16. 根据权利要求15所述的计算机设备，其特征在于，所述处理器还执行以下步骤：
    根据所述顶点像素生成所述验证码图像的第二边界；
    计算所述下一字符的起始边界与所述验证码图像的第二边界的距离；及
    当所述距离小于第二阈值时，则所述验证码图像中的所有字符识别完成。
  17. 根据权利要求13所述的计算机设备，其特征在于，所述处理器执行所述根据所述起始边界生成识别窗口之后，还执行以下步骤：
    当所述验证码图像的高度与所述识别窗口的高度不匹配时，计算所述验证码图像的高宽比；及
    根据所述验证码图像的高宽比调节所述验证码图像的高度和宽度。
  18. 根据权利要求13所述的计算机设备，其特征在于，所述处理器执行所述计算所述识别窗口中的当前字符为标准字符的概率之前，还包括：
    对所述验证码图像进行二值化处理；
    获取二值化处理后的验证码图像中的每个字符的边缘；及
    将获取到的每个字符的边缘进行平滑处理。
  19. 一种计算机存储介质，其上存储有计算机可读指令，其特征在于，该指令被处理器执行时实现以下步骤：
    获取验证码图像中当前字符的起始边界，并根据所述起始边界生成识别窗口，计算所述识别窗口中的当前字符为标准字符的概率；
    固定所述识别窗口的高度，且按照预设步长增加所述识别窗口的宽度，并计算宽度增加后的识别窗口中的字符为标准字符的概率，直至所述识别窗口的高宽比小于等于第一阈值；
    选取所计算的概率中的最大概率，并获取与所述最大概率对应的标准字符；及
    将所述最大概率对应的标准字符作为所述当前字符的识别结果输出。
  20. 根据权利要求19所述的计算机存储介质，其特征在于，所述一个或多个处理器执行的所述在将所述最大概率对应的标准字符作为所述当前字符的识别结果输出的步骤之后，还包括：
    根据所述最大概率对应的识别窗口的宽度以及所述当前字符的起始边界，计算下一字符的起始边界，并根据所述下一字符的起始边界识别所述验证码图像中的下一字符，直至所述验证码图像中的所有字符识别完成。
  21. 根据权利要求19所述的计算机存储介质，所述一个或多个处理器还执行以下步骤：
    识别所述验证码图像的边缘像素点，根据所述边缘像素点选取所述验证码图像的顶点像素点；及
    根据所述顶点像素生成所述验证码图像的第一边界，将所述第一边界作为所述验证码图像中的第一个字符的起始边界。
  22. 根据权利要求21所述的计算机存储介质，其特征在于，所述一个或多个处理器还执行以下步骤：
    根据所述顶点像素生成所述验证码图像的第二边界；
    计算所述下一字符的起始边界与所述验证码图像的第二边界的距离；及
    当所述距离小于第二阈值时，则所述验证码图像中的所有字符识别完成。
  23. 根据权利要求19所述的计算机存储介质，其特征在于，所述一个或多个处理器执行所述根据所述起始边界生成识别窗口之后，还包括：
    当所述验证码图像的高度与所述识别窗口的高度不匹配时，计算所述验证码图像的高宽比；及
    根据所述验证码图像的高宽比调节所述验证码图像的高度和宽度。
  24. 根据权利要求19所述的计算机存储介质，其特征在于，所述一个或多个处理器执行所述计算所述识别窗口中的当前字符为标准字符的概率之前，还包括：
    对所述验证码图像进行二值化处理；
    获取二值化处理后的验证码图像中的每个字符的边缘；及
    将获取到的每个字符的边缘进行平滑处理。
PCT/CN2017/104291 2017-08-28 2017-09-29 验证码识别方法、装置、计算机设备及计算机存储介质 WO2019041424A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710752752.7A CN107688809B (zh) 2017-08-28 2017-08-28 验证码识别方法、装置、计算机设备及计算机存储介质
CN201710752752.7 2017-08-28

Publications (1)

Publication Number Publication Date
WO2019041424A1 true WO2019041424A1 (zh) 2019-03-07

Family

ID=61155601

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/104291 WO2019041424A1 (zh) 2017-08-28 2017-09-29 验证码识别方法、装置、计算机设备及计算机存储介质

Country Status (2)

Country Link
CN (1) CN107688809B (zh)
WO (1) WO2019041424A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363195A (zh) * 2019-06-18 2019-10-22 深圳壹账通智能科技有限公司 图形验证码识别方法、装置、可读存储介质及终端设备
CN110766068A (zh) * 2019-10-21 2020-02-07 北京车智赢科技有限公司 一种验证码识别方法及计算设备
CN110866238A (zh) * 2019-11-13 2020-03-06 北京工业大学 基于对抗样本的验证码图像的生成方法
CN110991434A (zh) * 2019-11-19 2020-04-10 国网浙江省电力有限公司宁波供电公司 自助终端证件识别方法及装置
CN111966432A (zh) * 2020-06-30 2020-11-20 北京百度网讯科技有限公司 验证码处理方法、装置、电子设备以及存储介质
CN114723749A (zh) * 2022-06-06 2022-07-08 深圳市国人光速科技有限公司 提升喷印字符清晰度的图像处理方法、装置及喷印设备
CN115909019A (zh) * 2022-10-26 2023-04-04 吉林省吉林祥云信息技术有限公司 一种验证码图像识别的多模型节点场景中的调度方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110399894A (zh) * 2018-04-25 2019-11-01 百度在线网络技术（北京）有限公司 用于匹配图片中实体的方法和装置
CN109919160B (zh) * 2019-03-04 2021-03-23 深圳先进技术研究院 验证码识别方法、装置、终端及存储介质
CN111859354B (zh) * 2020-07-21 2023-09-01 百度在线网络技术（北京）有限公司 图片验证方法、装置、电子设备、存储介质和程序产品

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5963669A (en) * 1997-01-02 1999-10-05 Ncr Corporation Method of extracting relevant character information from gray scale image data for character recognition
CN104636747A (zh) * 2013-11-11 2015-05-20 北京大学 一种字符去粘连的方法及系统
CN105426890A (zh) * 2015-11-09 2016-03-23 成都数之联科技有限公司 一种字符扭曲粘连的图形验证码识别方法
CN105631449A (zh) * 2015-12-21 2016-06-01 华为技术有限公司 一种图片分割方法、装置及设备

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104252620B (zh) * 2014-09-25 2017-06-06 同程网络科技股份有限公司 字符粘连的图形验证码识别方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5963669A (en) * 1997-01-02 1999-10-05 Ncr Corporation Method of extracting relevant character information from gray scale image data for character recognition
CN104636747A (zh) * 2013-11-11 2015-05-20 北京大学 一种字符去粘连的方法及系统
CN105426890A (zh) * 2015-11-09 2016-03-23 成都数之联科技有限公司 一种字符扭曲粘连的图形验证码识别方法
CN105631449A (zh) * 2015-12-21 2016-06-01 华为技术有限公司 一种图片分割方法、装置及设备

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363195A (zh) * 2019-06-18 2019-10-22 深圳壹账通智能科技有限公司 图形验证码识别方法、装置、可读存储介质及终端设备
CN110766068A (zh) * 2019-10-21 2020-02-07 北京车智赢科技有限公司 一种验证码识别方法及计算设备
CN110766068B (zh) * 2019-10-21 2023-09-19 北京车智赢科技有限公司 一种验证码识别方法及计算设备
CN110866238A (zh) * 2019-11-13 2020-03-06 北京工业大学 基于对抗样本的验证码图像的生成方法
CN110866238B (zh) * 2019-11-13 2023-04-21 北京工业大学 基于对抗样本的验证码图像的生成方法
CN110991434A (zh) * 2019-11-19 2020-04-10 国网浙江省电力有限公司宁波供电公司 自助终端证件识别方法及装置
CN111966432A (zh) * 2020-06-30 2020-11-20 北京百度网讯科技有限公司 验证码处理方法、装置、电子设备以及存储介质
CN111966432B (zh) * 2020-06-30 2023-07-28 北京百度网讯科技有限公司 验证码处理方法、装置、电子设备以及存储介质
CN114723749A (zh) * 2022-06-06 2022-07-08 深圳市国人光速科技有限公司 提升喷印字符清晰度的图像处理方法、装置及喷印设备
CN114723749B (zh) * 2022-06-06 2022-10-11 深圳市国人光速科技有限公司 提升喷印字符清晰度的图像处理方法、装置及喷印设备
CN115909019A (zh) * 2022-10-26 2023-04-04 吉林省吉林祥云信息技术有限公司 一种验证码图像识别的多模型节点场景中的调度方法
CN115909019B (zh) * 2022-10-26 2024-02-09 吉林省吉林祥云信息技术有限公司 一种验证码图像识别的多模型节点场景中的调度方法

Also Published As

Publication number Publication date
CN107688809B (zh) 2019-08-27
CN107688809A (zh) 2018-02-13

Similar Documents

Publication Publication Date Title
WO2019041424A1 (zh) 验证码识别方法、装置、计算机设备及计算机存储介质
US12073542B2 (en) Image processing method, image processing apparatus, and non-transitory storage medium
US11113836B2 (en) Object detection method, device, apparatus and computer-readable storage medium
US10803338B2 (en) Method and device for recognizing the character area in a image
US10929648B2 (en) Apparatus and method for data processing
WO2021056746A1 (zh) 图像模型检测方法、装置、电子设备及存储介质
TW202011733A (zh) 對影像進行目標取樣的方法及裝置
US10282604B2 (en) Automating creation of accurate OCR training data using specialized UI application
US10410053B2 (en) Method, apparatus, system, and storage medium for detecting information card in image
US9082181B2 (en) Image processing method and apparatus
CN111259878A (zh) 一种检测文本的方法和设备
US10025976B1 (en) Data normalization for handwriting recognition
RU2641225C2 (ru) Способ выявления необходимости обучения эталона при верификации распознанного текста
US9977976B2 (en) Path score calculating method for intelligent character recognition
US10169673B2 (en) Region-of-interest detection apparatus, region-of-interest detection method, and recording medium
CN113657202A (zh) 构件识别方法、构建训练集方法、装置、设备及存储介质
CN112001406A (zh) 一种文本区域检测方法及装置
WO2021196013A1 (zh) 单词识别方法、设备及存储介质
CN115359008A (zh) 显示界面的测试方法及装置、存储介质、电子设备
US11238620B2 (en) Implicit structured light decoding method, computer equipment and readable storage medium
US9684844B1 (en) Method and apparatus for normalizing character included in an image
CN111368674B (zh) 图像识别方法及装置
RU2458396C1 (ru) Способ редактирования статических цифровых комбинированных изображений, включающих в себя изображения нескольких объектов
US9152876B1 (en) Methods and systems for efficient handwritten character segmentation
KR101660596B1 (ko) 안면 형상 기울기 보정 방법 및 보정 시스템

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: 17923725

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 07/09/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17923725

Country of ref document: EP

Kind code of ref document: A1