JP6344389B2 - Optical character recognition device - Google Patents

Description

  The present invention relates to an optical character recognition device that optically recognizes a character string, and more particularly to an optical character recognition device that recognizes a character string representing a date. The present invention also relates to an optical character recognition method, a computer program, and a recording medium for recognizing a character string representing a date.

  There is a need for a device that optically recognizes characters printed on a medicine container (see Patent Document 1). For example, if a medicine such as an injection that has been transported to the ward and is not used as a result is returned to the vault, the medicine type, name, and so that the medicine can be quickly and definitely removed at the next use. It is necessary to sort and store based on the expiration date. If a return device that automatically performs this sorting can be realized using an optical character recognition device, it is effective for improving work efficiency and reducing errors. In addition, when storing medicine using such a return device, the storage location of the medicine is recorded, so that the appropriate medicine can be automatically dispensed based on the prescription at the next use. .

Japanese Patent No. 4857768.

  When a character string is optically recognized by the optical character recognition device, erroneous recognition may occur due to various noises included in the character string image. In order to improve the recognition accuracy, it is necessary to remove noise from the image in advance.

  An object of the present invention is to provide an optical character recognition device, an optical character recognition method, a computer program, and a recording medium capable of recognizing a character string representing a date with higher accuracy than the conventional one. .

According to the optical character recognition device according to the first aspect of the present invention, in the optical character recognition device that optically recognizes a character string, the optical character recognition device includes:
First processing means for extracting a target area including an object to be recognized from an input image;
Second processing means for extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. And third processing means for recognizing column candidates as dates.

According to the optical character recognition device according to the second aspect of the present invention, in the optical character recognition device according to the first aspect,
The second processing means includes
Detect the contour and edge of the object included in the target area,
An object having contours and edges that overlap each other is extracted as the candidate object.

According to the optical character recognition device according to the third aspect of the present invention, in the optical character recognition device according to the second aspect,
The second processing means includes
Applying a Sobel filter to the target area to detect the first edge;
Applying a canny filter to a region in the vicinity of the first edge to detect a second edge;
The second edge is used as an edge of an object included in the target area.

According to the optical character recognition device according to the fourth aspect of the present invention, in the optical character recognition device according to any one of the first to third aspects,
The third processing means includes
Label the above character string candidate objects to extract multiple character candidates,
A plurality of bounding boxes each having a rectangular shape having a width parallel to a direction in which the character string candidates extend and a height orthogonal to the direction in which the character string candidates extend, and surrounding each of the character candidates. Generate and
Each of the bounding boxes is deformed so that the width of the bounding box is increased as the height of the bounding box is lower.
A set of character candidates included in bounding boxes connected by deformation is extracted as a new character string candidate.

According to the optical character recognition device according to the fifth aspect of the present invention, in the optical character recognition device according to any one of the first to fourth aspects,
The third processing means includes
Label the above character string candidate objects to extract multiple character candidates,
Character string candidates including more than 10 character candidates are deleted.

According to the optical character recognition device according to the sixth aspect of the present invention, in the optical character recognition device according to any one of the first to fifth aspects,
The third processing means includes
Label the above character string candidate objects to extract multiple character candidates,
Character string candidates including only character candidates including two or more objects in a direction orthogonal to the direction in which the character string candidates extend are deleted.

According to the optical character recognition device according to the seventh aspect of the present invention, in the optical character recognition device according to any one of the first to sixth aspects,
The third processing means includes
Detect the contours and edges of the above character string candidate objects,
Character string candidates whose portion where the edge pixel coincides with the contour pixel are 60% or less of the area of the edge pixel are deleted.

According to the optical character recognition device according to the eighth aspect of the present invention, in the optical character recognition device according to any one of the first to seventh aspects,
The third processing means recognizes the character string candidate as not being a date when the character string candidate includes an alphabetic character that is clearly not mistaken for a number.

According to the optical character recognition device according to the ninth aspect of the present invention, in the optical character recognition device according to any one of the first to eighth aspects,
When the character string candidate includes two numbers representing the month and at least one other character subsequent to the two numbers representing the month, the third processing unit is configured to represent the month. When the distance between two numbers is larger than the average of the distance between the two numbers representing the month and other characters and the average distance between the other characters, the first digit of the two numbers representing the month and The other characters are removed.

According to the optical character recognition device according to the tenth aspect of the present invention, in the optical character recognition device according to any one of the first to ninth aspects,
The input image is an image of a cylindrical container held rotatably.

According to the optical character recognition device in the eleventh aspect of the present invention, in the optical character recognition device in the tenth aspect,
The first processing means includes, from the input image, an edge extending in a direction substantially perpendicular to a rotation axis of the cylindrical container, and a portion having a luminance higher than a predetermined threshold value. A region to be included is extracted as the target region.

According to the optical character recognition device in the twelfth aspect of the present invention, in the optical character recognition device in the tenth or eleventh aspect,
The optical character recognition device acquires a plurality of input images respectively representing different angles of the container photographed while rotating the container,
When the character string candidate related to one input image includes only “1” as the number representing the month, the third processing means only includes “1” as the number representing the month as the character string candidate related to the other input image. It is characterized by determining whether it is included.

According to the optical character recognition device according to the thirteenth aspect of the present invention, in the optical character recognition device according to any one of the tenth to twelfth aspects,
The optical character recognition device is
A camera,
An imaging table for holding the container so as to be rotatable around a rotation axis of the cylindrical container;
A moving device for moving the container between at least one storage and the imaging table;
The container is printed with a character string representing the expiration date of the medicine in the container.

According to the optical character recognition method of the fourteenth aspect of the present invention,
In the optical character recognition method for optically recognizing a character string, the optical character recognition method includes:
A first step of extracting a target area including an object to be recognized from an input image;
A second step of extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. And a third step of recognizing column candidates as dates.

According to a computer program according to the fifteenth aspect of the present invention,
In a computer program that optically recognizes a character string when executed by a computer, the computer program includes:
A first step of extracting a target area including an object to be recognized from an input image;
A second step of extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. And a third step of recognizing column candidates as dates.

According to the recording medium of the sixteenth aspect of the present invention,
In a computer-readable recording medium storing a computer program for optically recognizing a character string when executed by a computer, the computer program includes:
A first step of extracting a target area including an object to be recognized from an input image;
A second step of extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. And a third step of recognizing column candidates as dates.

According to the optical character recognition device according to the seventeenth aspect of the present invention, in the optical character recognition device according to any one of the first to thirteenth aspects,
The optical character recognition device is characterized in that it acquires a plurality of input images each representing a different angle of the container taken while rotating the container, and connects the plurality of input images.

  The optical character recognition device, the optical character recognition method, the computer program, and the recording medium of the present invention can recognize a character string representing a date with higher accuracy than before.

It is a block diagram which shows the structure of the optical character recognition apparatus which concerns on the 1st Embodiment of this invention. It is a top view which shows the container 13a on which the character string which concerns on a 1st example was printed. It is a top view which shows the container 13b on which the character string which concerns on a 2nd example was printed. It is a top view which shows the container 13c on which the character string which concerns on a 3rd example was printed. It is a top view which shows the container 13d on which the character string which concerns on a 4th example was printed. It is a flowchart which shows the date detection process performed by the control apparatus 1 of FIG. It is a flowchart which shows the subroutine of the target area | region extraction process in FIG.6 S2. It is a flowchart which shows the subroutine of the candidate object extraction process in step S4 of FIG. It is a flowchart which shows the subroutine of the OCR process in step S6 of FIG. 10 is a flowchart showing a first part of an OCR subroutine in steps S51, S53, S55, and S57 of FIG. 10 is a flowchart showing a second part of the OCR subroutine in steps S51, S53, S55, and S57 of FIG. 12 is a flowchart showing a subroutine of edge strength and area luminance determination processing in step S68 of FIG. It is a flowchart which shows the subroutine of the average height determination process in step S69 of FIG. It is a flowchart which shows the subroutine of the date pattern determination process in step S72, S75 of FIG. It is a figure which shows the example of the part with high brightness | luminance in the image extracted in step S21 of FIG. It is a figure which shows the example of the long vertical edge in the image extracted in step S22 of FIG. FIG. 17 is a diagram in which a portion with high luminance in FIG. 15 and a long vertical edge in FIG. 16 are superimposed. It is a figure which shows the target area | region 21 containing the high brightness | luminance part of FIG. 15, and the long vertical edge of FIG. It is a figure which shows the example of the bright object extracted in step S31 of FIG. It is a figure which shows the example of the dark object extracted in step S33 of FIG. It is a figure which shows extraction of the bright object using the moving average filter in step S31 of FIG. It is a figure which shows the image containing a bright object and a dark object. It is a figure which shows extraction of the bright object from the image of FIG. 22A. It is a figure which shows extraction of the dark object from the image of FIG. 22A. It is a figure which shows the example of the edge in the image extracted using the Sobel filter in step S35 of FIG. It is a figure which shows the example of the image after deleting area | regions other than the edge extracted in step S35 in step S36 of FIG. It is a figure which shows the example of the edge in the image extracted using the canny filter of the threshold value 15 in step S37 of FIG. It is a figure which shows the example of the edge in the image extracted using the canny filter of the threshold value 4 in step S37 of FIG. It is a figure which shows the example of the candidate object of the bright object extracted in step S38 of FIG. It is a figure which shows the example of the candidate object of the dark object extracted in step S39 of FIG. It is a figure which shows the example of an object. FIG. 29B is a diagram showing the binarized image of FIG. 29A extracted using the threshold value 200. It is a figure which shows the outline of FIG. 29B. It is a figure which shows the edge of FIG. 29A extracted using the threshold value 50. FIG. It is a figure which shows the edge of FIG. 29A extracted using the threshold value 200. FIG. It is a figure which shows the outline and edge of a character. It is a figure which shows the example of the character string candidate extracted in step S61 of FIG. It is a figure which shows extraction of the character string candidate in step S61 of FIG. It is a figure which shows the example of the character candidate extracted in step S62 of FIG. 10, and the produced | generated bounding box. It is a figure which shows the example of the character string candidate extracted in step S61 of FIG. It is a figure which shows the example of the character candidate extracted in step S62 of FIG. 10, and the produced | generated bounding box 42. FIG. It is a figure which shows the example of the bounding box 43 deform | transformed in FIG.10 S63. It is a figure which shows the example of the new character string candidate extracted in step S64 of FIG. It is a figure which shows the other example of the character string candidate extracted in step S61 of FIG. It is a figure which shows the other example of the character candidate extracted in step S62 of FIG. 10, and the generated bounding box 42. FIG. It is a figure which shows the example of the character string candidate after deleting some character string candidates in step S65, S66, S67 of FIG. It is a figure which shows the example of the character string candidate extracted in step S64 of FIG. It is a figure which shows the number of the objects in the height direction of each character candidate contained in the character string candidate of FIG. 36A. It is a figure which shows the other example of the character string candidate extracted in step S64 of FIG. It is a figure which shows the number of objects in the height direction of each character candidate contained in the character string candidate of FIG. 37A. It is a figure which shows the example of an input image. It is a figure which shows the candidate object of the bright object extracted from the image of FIG. 38A. It is a figure which shows the outline of the candidate object of FIG. 38B. It is a figure which shows the edge extracted from the image of FIG. 38A. It is a figure which shows the example of an input image. It is a figure which shows the candidate object of the dark object extracted from the image of FIG. 39A. It is a figure which shows the outline of the candidate object of FIG. 39B. It is a figure which shows the edge extracted from the image of FIG. 39A. It is a figure which shows the example of the input image for demonstrating the edge intensity | strength and area | region brightness | luminance determination processing in FIG.11 S68. It is a figure which shows the example of the character string candidate selected in step S90 of FIG. It is a figure which shows the example of the character string candidate processed by the edge intensity | strength and area | region brightness | luminance determination process in FIG.11 S68. It is a figure which shows the example of the character string candidate selected in step S101 of FIG. It is a figure which shows the example of the new character string candidate extracted in step S104 of FIG. It is a flowchart which shows the date detection process performed by the control apparatus 1 of the optical character recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a flowchart which shows the date detection process performed by the control apparatus 1 of the optical character recognition apparatus which concerns on the 3rd Embodiment of this invention. It is a flowchart which shows the subroutine of the date detection process of the connection image in step S15 of FIG. It is a flowchart which shows the subroutine of the date pattern determination process of the date detection process performed by the control apparatus 1 of the optical character recognition apparatus which concerns on the 4th Embodiment of this invention. It is a figure which shows the example of the character string candidate containing the character string of a date, and another character.

First embodiment.
FIG. 1 is a block diagram showing the configuration of the optical character recognition apparatus according to the first embodiment of the present invention. The optical character recognition device in FIG. 1 optically recognizes a date character string printed on the surface of a cylindrical container 13.

  The optical character recognition device in FIG. 1 includes a control device 1, a rail 2, a moving device 3, cameras 4 to 6, illumination devices 7a and 7b, and rollers 8a and 8b. The at least two rollers 8a and 8b are provided in parallel with each other and are provided with a driving device that operates under the control of the control device 1, thereby holding the container 13 rotatably. The optical character recognition device further includes at least one tray (or storage) 11 and 12 that accommodates the container 13. The moving device 3 moves the container 13 between the trays 11 and 12 and the rollers 8 a and 8 b under the control of the control device 1. The cameras 4 to 6 are provided on the trays 11 and 12 and the rollers 8a and 8b, respectively, and acquire images of the containers 13 when the containers 13 are on the trays 11 and 12 and the rollers 8a and 8b. Send to. The lighting devices 7a and 7b illuminate the container 13 on the rollers 8a and 8b. The rollers 8 a and 8 b and the lighting devices 7 a and 7 b function as a photographing stand for the container 13. The optical character recognition device may include another mechanism that holds the container 13 so as to be rotatable around the rotation axis of the cylindrical container 13 instead of the rollers 8a and 8b. The control device 1 performs date detection processing described later with reference to FIGS. 6 to 14 on the image of the container 13 sent from the camera 5 to recognize the date printed on the surface of the container 13. An additional camera may be provided in the moving device 3. The control device 1 may be connected to an external personal computer (PC) 9 that operates according to a computer program read from the recording medium 10.

  The container 13 is, for example, a medicine container (ampoule), and a character string representing the expiration date of the medicine in the container 13 is printed on the container 13. For example, when such a container 13 is returned from the ward and placed on the tray 11, the optical character recognition device moves the container 13 from the tray 11 onto the rollers 8 a and 8 b using the moving device 3, and the roller 8 a , 8b, the expiration date printed on the container 13 is optically recognized. The optical character recognition device then determines, based on the recognized date, whether to store or discard the container and uses the transfer device 3 to transfer to another tray 12 associated with the appropriate storage or trash. The container 13 is moved.

  2 to 5 show examples of character strings printed on the container 13. FIG. 2 is a top view showing the container 13a on which the character string according to the first example is printed. FIG. 3 is a top view showing the container 13b on which the character string according to the second example is printed. FIG. 4 is a top view showing the container 13c on which the character string according to the third example is printed. FIG. 5 is a top view showing a container 13d on which a character string according to the fourth example is printed. The character string may be printed on a label attached to the container, or may be printed directly on the container. Further, the direction of the character string may be parallel to the rotation axis of the cylindrical container 13, may be orthogonal to the rotation axis of the container 13, or the character strings in these directions may be mixed. Good.

  The control device 1 operates as a first processing unit that extracts a target area including an object to be recognized from an input image. In addition, the control device 1 operates as a second processing unit that extracts candidate objects including at least one character string candidate object from the objects included in the target area. In addition, the control device 1 performs labeling of the candidate objects, extracts a plurality of objects extending in a predetermined direction and close to each other as character string candidates, and the character string candidates are two digits representing the year or Judge whether or not it has a date pattern that includes a 4-digit number, a 1- or 2-digit number representing the month, and a predetermined punctuation mark. Operates as a third processing means for recognizing a character string candidate as a date.

  Hereinafter, the date detection process executed by the control device 1 of FIG. 1 will be described with reference to FIGS.

  FIG. 6 is a flowchart showing date detection processing executed by the control device 1 of FIG. The control device 1 takes images of the container 13 with the camera 5 while rotating the container 13 by a certain angle (for example, 15 degrees) using the rollers 8a and 8b, and displays a plurality of images (input images) representing different angles of the container 13 respectively. ) To get. As the camera 5, a camera having sufficient resolution for optically recognizing the character string printed on the container 13 is used. For example, the container 13 has a diameter of 10 to 40 mm. For example, a black and white camera that captures a range of 120 × 90 mm including the container 13 with a pixel number of 3840 × 2748 (about 10 million pixels) is used. In this case, 1 mm on the container 13 corresponds to 32 pixels. In step S <b> 1 of FIG. 6, the control device 1 acquires one of the plurality of images of the container 13. In step S2, the control device 1 executes target area extraction processing.

  FIG. 7 is a flowchart showing a subroutine of target area extraction processing in step S2 of FIG.

  In step S21 of FIG. 7, the control device 1 extracts a portion having a luminance higher than a predetermined threshold (for example, a portion including illumination reflection) from the image acquired in step S1. For example, when the luminance of the pixel changes from 0 to 255, the control device 1 extracts a portion having a luminance higher than 220, for example. FIG. 15 is a diagram illustrating an example of a portion having a high luminance in the image extracted in step S21 of FIG. Here, the input image is an image of the container 13d in FIG.

  Next, in step S22 of FIG. 7, the control device 1 extracts a long edge (longitudinal edge) extending in a direction substantially perpendicular to the rotation axis of the cylindrical container 13 from the image acquired in step S1. To do. Although the rollers 8a and 8b exist in the background of the container 13, since the rollers 8a and 8b extend in parallel with the rotation axis of the container 13, the influence of the rollers 8a and 8b is removed by extracting the vertical edge. be able to.

  In order to extract an edge in step S22, the following Sobel filter is used.

  Among the edges extracted by the Sobel filter, short edges, for example, edges shorter than 55 pixels are deleted as noise. FIG. 16 is a diagram illustrating an example of a long vertical edge in the image extracted in step S22 of FIG. The rotation axis of the container 13d in FIG. 5 is parallel to the X axis in FIG. 5. Therefore, in step S22, an edge extending in a direction substantially parallel to the Y axis in FIG. 5 is extracted.

  In step S23 of FIG. 7, the control device 1 extracts a rectangular area (width w1 × height h1) including a portion with high brightness and a vertical edge as a target area, and deletes an area outside the target area. FIG. 17 is a diagram in which the high luminance portion of FIG. 15 and the long vertical edge of FIG. 16 are superimposed. FIG. 18 is a diagram illustrating the target area 21 including the high-luminance portion of FIG. 15 and the long vertical edge of FIG. The target area is an area that is considered to include a character string object to be recognized.

  Referring to FIG. 6 again, after executing the target area extraction process in step S2, in step S3, the control device 1 determines whether or not the target area has been successfully extracted. If YES, the process proceeds to step S4. If NO, the process proceeds to step S10. In step S4, the control device 1 executes candidate object extraction processing.

  FIG. 8 is a flowchart showing a subroutine of candidate object extraction processing in step S4 of FIG.

  In step S31 of FIG. 8, the control device 1 applies a moving average filter to the image of the target area, thereby extracting bright objects brighter than the surroundings and binarizing the image. A white character string on a black background is extracted as a bright object. Since lighting is uneven, it is not possible to detect an object by simply performing binarization. Therefore, a binarization method (dynamic threshold method) using a moving average filter is used. FIG. 21 is a diagram illustrating extraction of a bright object using the moving average filter in step S31 of FIG. According to the principle shown in FIG. 21, the control device 1 calculates the local average luminance from the luminance of the input image (here, the image of the target area), and adds a predetermined offset to the local average luminance. An object having a brightness higher than the brightness (that is, an object having a brightness that protrudes from the surroundings) is extracted as a bright object. The size of the local area that is referred to for calculating the local average luminance is determined based on the size of the entire target area. FIG. 19 is a diagram illustrating an example of the bright object extracted in step S31 of FIG. Next, in step S32, the control device 1 detects the contour of the binarized bright object.

  Next, in steps S33 to S34 of FIG. 8, the same processing as that performed on the bright object in steps S31 to S32 is performed on the dark object. In step S33, the control device 1 applies a moving average filter to the image of the target area, thereby extracting dark objects that are darker than the surroundings and binarizing the image. A black character string on a white background is extracted as a dark object. FIG. 20 is a diagram illustrating an example of the dark object extracted in step S33 of FIG. In step S34, the control device 1 detects the binarized dark object contour.

  FIG. 22A is a diagram illustrating an image including a bright object and a dark object. FIG. 22B is a diagram showing extraction of a bright object from the image of FIG. 22A. FIG. 22C is a diagram illustrating extraction of a dark object from the image of FIG. 22A. One character string is considered to be one of a bright object and a dark object. Since both the bright object and the dark object are extracted in steps S31 to S34 in FIG. 8, the date printed on the container 13 can be reliably detected.

  Next, in step S35 of FIG. 8, the control device 1 extracts an edge in the image of the target area by using the following Sobel filter.

  Here, b indicates the result of applying the operator B to a certain pixel, and c indicates the result of applying the operator C to the same pixel. FIG. 23 is a diagram showing an example of edges in the image extracted using the Sobel filter in step S35 of FIG.

  Next, in step S36 of FIG. 8, the control device 1 deletes an area other than the edge extracted in step S35 from the image of the target area. FIG. 24 is a diagram illustrating an example of an image after deleting a region other than the edge extracted in step S35 in step S36 of FIG. Next, in step S37 in FIG. 8, the control device 1 applies a Canny filter to the image after the deletion in step S36, and extracts an edge in the image.

  The Canny edge detection method includes the following three steps. As a first step, the gradient magnitude g (x, y) and gradient direction d (x, y) of the following equation are calculated for the image.

Here, f _x (x, y) indicates a convolution of the first-order derivative in the x direction of the Gaussian function having the standard deviation σ and the pixel value function, and f _y (x, y) is y of the same Gaussian function. The first derivative with respect to the direction and the convolution of the pixel value function are shown.

  As a second step of the Canny edge detection method, an edge is detected by obtaining the maximum value of the gradient magnitude g (x, y). At this time, the magnitude of the gradient of the pixel of interest is estimated by estimating the magnitude of the gradient interpolated with respect to the gradient direction d (x, y) using 8 pixels around the pixel of interest, and comparing with the estimated value. It is determined whether g (x, y) has a true maximum value.

  As a third step of the Canny edge detection method, a high threshold value Th_H and a low threshold value Th_L are set, and a threshold value with hysteresis is determined. When the gradient magnitude g (x, y) is greater than the high threshold Th_H, the pixel is determined to be an edge. When the gradient magnitude g (x, y) is smaller than the low threshold Th_L, it is determined that the pixel is not an edge. When the gradient magnitude g (x, y) is between the high threshold Th_H and the low threshold Th_L, it is determined that the pixel is an edge only when the pixel is adjacent to a pixel detected as an edge. To do.

  In the example of the present disclosure, a canny filter having a Gaussian function standard deviation σ = 1.4, a high threshold Th_H = 10, and a low threshold Th_L = 5 is used in step S37. The range of possible values for the threshold is 0-255.

  In step 40 of FIG. 8, two types of edges are extracted using two types of canny filters for use in both steps S38 and S39 immediately after that and step S67 of FIG. 10 described later. FIG. 25 is a diagram illustrating an example of edges in the image extracted using the Canny filter having the threshold value 15 in step S37 of FIG. FIG. 26 is a diagram illustrating an example of edges in the image extracted using the Canny filter having the threshold value 4 in step S37 of FIG. In the example of the present disclosure, the edge of FIG. 26 is used in steps S38 and S39, and the edge of FIG. 25 is used in step S67 of FIG.

  Extracting an edge using a Sobel filter is fast, but the width of the extracted edge increases. Extracting edges using the Canny filter can extract detailed edges at a low speed. On the other hand, in this embodiment, edges are extracted once using a Sobel filter (step S35), areas other than the extracted edges are deleted (step S36), and the vicinity of the edges extracted using the Sobel filter is used. An edge is extracted by applying a canny filter only to the area (step S37), and this edge is used as an edge of an object included in the target area. Thus, by combining the Sobel filter and the Canny filter, the edge extraction is about 10 times faster than when only the Canny filter is used.

  Next, in step S38 in FIG. 8, the control device 1 extracts a bright object having a contour and an edge, the contour and the edge overlapping each other and substantially matching each other as a candidate object. FIG. 27 is a diagram illustrating an example of a bright object candidate object extracted in step S38 of FIG. Next, in step S39 in FIG. 8, the control device 1 extracts dark objects having contours and edges, the contours and edges overlapping each other and substantially matching each other as candidate objects. FIG. 28 is a diagram illustrating an example of dark object candidate objects extracted in step S39 of FIG.

  FIG. 29A shows an example of an object. FIG. 29B is a diagram showing the binarized image of FIG. 29A extracted using the threshold value 200. FIG. 29C is a diagram showing an outline of FIG. 29B. FIG. 29D is a diagram showing the edges of FIG. 29A extracted using the threshold value 50. FIG. 29E is a diagram showing the edges of FIG. 29A extracted using the threshold 200. FIG. 29F is a diagram showing the outline and edges of a character. The object in FIG. 29A includes, for example, when the pixel luminance varies from 0 to 255, a luminance 0 portion, a luminance 128 portion, and a luminance 255 portion. The outline of the object in FIG. 29A is obtained as the outline of the binarized image (FIG. 29B) (FIG. 29C). The edge of the object in FIG. 29A is obtained as a portion where the luminance changes suddenly, and different edges are extracted by using different threshold values (FIGS. 29D and 29E). As shown in FIGS. 29C to 29E, generally, the outline and edge of an object do not always match. However, it is considered that the character object always has a closed edge, and the outline and the edge of the object coincide. Therefore, a character object can be extracted by extracting an object having a substantially matched contour and edge. Objects whose edges and contours do not match are deleted as noise.

  Next, in step S5 of FIG. 6, the control device 1 determines whether or not the candidate object has been successfully extracted. If YES, the process proceeds to step S6, and if NO, the process proceeds to step S10. In step S6, the control device 1 executes an OCR process.

  FIG. 9 is a flowchart showing a subroutine of OCR processing in step S6 of FIG. It is unknown whether the character string to be recognized is a bright object or a dark object, and whether the character string to be recognized extends parallel to the X axis in FIG. 5 or extends parallel to the Y axis. Since it is unknown whether it exists, the OCR subroutine of FIGS. 10 and 11 is executed for all these combinations. When it is assumed that the character string of the recognition device is a bright object, the bright object candidate object extracted in step S38 in FIG. 8 is used. When it is assumed that the character string of the recognition device is a dark object, the dark object candidate object extracted in step S39 in FIG. 8 is used. When it is assumed that the character string of the recognition device extends parallel to the X axis, the image of the target area is used as it is. When it is assumed that the character string of the recognition device extends parallel to the Y axis, the image of the target area is rotated 90 degrees and used.

  In step S51 of FIG. 9, the control device 1 executes the OCR subroutine on the assumption that the character string of the recognition device is a bright object extending parallel to the X axis. In step S52, the control device 1 determines whether or not the OCR is successful. If YES, the process proceeds to step S7 in FIG. 6, and if NO, the process proceeds to step S53. In step S53, the control device 1 executes the OCR subroutine on the assumption that the character string of the recognition device is a bright object extending in parallel with the Y axis. In step S54, the control device 1 determines whether or not the OCR is successful. If YES, the process proceeds to step S7 in FIG. 6, and if NO, the process proceeds to step S55. In step S55, the control device 1 executes the OCR subroutine on the assumption that the character string of the recognition device is a dark object extending in parallel with the X axis. In step S56, the control device 1 determines whether or not the OCR is successful. If YES, the process proceeds to step S7 in FIG. 6, and if NO, the process proceeds to step S57. In step S57, the control device 1 executes the OCR subroutine on the assumption that the character string of the recognition device is a dark object extending in parallel to the Y axis, and then proceeds to step S7 of FIG.

  FIG. 10 is a flowchart showing a first part of the OCR subroutine in steps S51, S53, S55, and S57 of FIG. FIG. 11 is a flowchart showing a second part of the OCR subroutine in steps S51, S53, S55, and S57 of FIG.

  In step S61 of FIG. 10, the control device 1 performs labeling of candidate objects, and extracts a plurality of objects extending in a predetermined direction and close to each other as character string candidates. FIG. 30 is a diagram showing an example of character string candidates extracted in step S61 of FIG. FIG. 31 is a diagram showing extraction of character string candidates in step S61 of FIG. In FIG. 31, a character string candidate mask 31 for extracting character string candidates has, for example, a width w2 = 75 pixels and a height h2 = 3 pixels. When the character string candidate mask 31 arranged at a certain position includes at least one candidate object pixel, the region in the character string candidate mask 31 is determined to be a part of the character string candidate. The character string candidate mask 31 is scanned over the entire target area, and labels are assigned to the individual connected character string candidates.

  When labeling candidate objects and extracting character string candidates in step S61, a plurality of adjacent character strings may be extracted as one character string candidate. Therefore, the character string candidates are once separated into character candidates, and character candidates having similar feature values are recombined as character string candidates based on the feature values (width and height) of each character candidate. In step S62 of FIG. 10, the control device 1 performs labeling of objects in each character string candidate, extracts a plurality of character candidates included in the character string candidate, and generates a bounding box for each character candidate. Each bounding box has a rectangular shape having a width parallel to the direction in which the character string candidates extend and a height orthogonal to the direction in which the character string candidates extend, and encloses each character candidate. It is. In step S63, the control device 1 deforms each bounding box so that the width of the bounding box is increased as the height of the bounding box is lower, based on the width and height. In step S64, the control device 1 extracts a set of character candidates included in the bounding boxes connected by deformation as a new character string candidate. FIG. 32 is a diagram illustrating an example of the character candidates extracted in step S62 of FIG. 10 and the generated bounding box.

  FIG. 33A is a diagram showing an example of character string candidates extracted in step S61 of FIG. FIG. 33B is a diagram showing an example of the character candidates extracted in step S62 of FIG. 10 and the generated bounding box 42. FIG. 33C is a diagram illustrating an example of the bounding box 43 modified in step S63 of FIG. FIG. 33D is a diagram showing an example of a new character string candidate extracted in step S64 of FIG. The character string candidates in FIG. 33A include two character strings “2012. 1” and “abc”, but are extracted as one character string candidate. FIG. 33A shows a bounding box 41 of character string candidates for explanation. Next, as shown in FIG. 33B, the objects in the character string candidates in FIG. 33A are labeled to extract a plurality of character candidates included in the character string candidates, and a bounding box 42 for each character candidate is generated. Each character candidate's bounding box 42 has a width w3 and a height h3. Next, as shown in FIG. 33C, each bounding box is deformed based on its width and height. The width w3 'and the height h3' of the bounding box 43 after deformation are obtained by the following equations.

  Here, W is the maximum value of the width of the bounding box 42 of each character candidate, and H is the maximum value of the height of the bounding box 42 of each character candidate.

  As shown in FIG. 33C, the bounding box of each character candidate is deformed so that the width w3 is increased as the height h3 is lower. Therefore, although the distance between “.” And “1” is larger than the distance between “1” and “a”, “.” And “1” are connected in the bounding box 43 after deformation, and “1” and “1” a ”is separated. As shown in FIG. 33D, a set of character candidates included in the bounding boxes connected by deformation is extracted as a new character string candidate. FIG. 33D shows new character string candidate bounding boxes 41a and 41b for explanation.

  FIG. 34A is a diagram showing another example of the character string candidates extracted in step S61 of FIG. FIG. 34B is a diagram showing another example of the character candidates extracted in step S62 of FIG. 10 and the generated bounding box 42. In step S64, when the set of character candidates included in the bounding boxes connected by deformation is not extracted as a new character string candidate, the object included in the character string candidate in FIG. 34A is deleted as noise.

  Next, in steps S65 to S67, character string candidates that are clearly not date character strings are deleted to prevent erroneous recognition.

  In step S65 of FIG. 10, the control device 1 deletes character string candidates including more than ten character candidates. The date string is considered to contain no more than 10 characters. Accordingly, character string candidates including more than 10 character candidates are deleted as noise.

  Next, in step S66, the control device 1 deletes a character string candidate including only a character candidate including two or more objects in the height direction. Here, in each character candidate, the objects included in the character candidate are connected in the height direction, and the number of connected objects is counted. The numbers “0” to “9” are single connected objects. Therefore, if the character string candidate is a date, all the character candidates included in the character string candidate should include only one object in the height direction. However, there is a possibility that there is a character candidate including two or more objects in the height direction due to noise etc. even though it is a date character string (if there are extra objects, the connected objects will be cut In such a case, character string candidates including only character candidates including two or more objects in the height direction are deleted as noise. FIG. 36A is a diagram showing an example of character string candidates extracted in step S64 of FIG. FIG. 36B is a diagram illustrating the number of objects in the height direction of each character candidate included in the character string candidates in FIG. 36A. The number of objects in the height direction of each character candidate is shown on the character candidate 51 in FIG. 36B.

  FIG. 37A is a diagram showing another example of the character string candidates extracted in step S64 of FIG. FIG. 37B is a diagram illustrating the number of objects in the height direction of each character candidate included in the character string candidates in FIG. 37A. FIG. 37A and FIG. 37B are examples in which a character string is processed as being erroneously extended horizontally even though it extends vertically. As shown in FIG. 37B, since the number of objects in the height direction of each character candidate is two or more, the character string candidates in FIGS. 37A and 37B are deleted as noise.

  Next, in step S67, the control device 1 clears a character string candidate in which a portion where the edge pixel in each object matches the contour pixel is 60% or less of the edge pixel area (number of pixels) from a certain region. When candidate objects of the object and the dark object are extracted, a correct character string candidate is included in only one of them. In a correctly extracted candidate object, it is considered that the edge pixel substantially matches the contour pixel. On the other hand, the character string candidate extracted from the other candidate object is deleted as noise. The edge extracted in step S37 in FIG. 8 is used as the object edge. FIG. 38A is a diagram illustrating an example of an input image. FIG. 38B is a diagram showing bright object candidate objects extracted from the image of FIG. 38A. FIG. 38C is a diagram showing an outline of the candidate object of FIG. 38B. FIG. 38D is a diagram showing edges extracted from the image of FIG. 38A. FIG. 39A is a diagram illustrating an example of an input image. FIG. 39B is a diagram showing dark object candidate objects extracted from the image of FIG. 39A. FIG. 39C is a diagram showing an outline of the candidate object in FIG. 39B. FIG. 39D is a diagram showing edges extracted from the image of FIG. 39A. The input images in FIGS. 38A and 39A include dark objects. Accordingly, the contour of FIG. 39C and the edge of FIG. 39D substantially match, but the contour of FIG. 38C and FIG. 38D do not match.

  FIG. 35 is a diagram showing an example of character string candidates after deleting some character string candidates in steps S65, S66, and S67 of FIG. Compared to FIG. 30, it can be seen that noise is reduced.

  Next, in step S68 of FIG. 11, the control device 1 executes edge strength and area luminance determination processing.

  FIG. 12 is a flowchart showing a subroutine of the edge strength and area brightness determination processing in step S68 of FIG.

  In step S90 of FIG. 12, the control device 1 selects one character string candidate. In step S91, the control device 1 extracts and expands the outline of the character candidate region. Here, the pixel is expanded by adding one pixel to the outline pixel of the character candidate area. In step S92, the control device 1 applies a canny filter to the contour expanded in step S91 to detect the edge of the character string candidate region. Here, a Canny filter having a Gaussian function standard deviation σ = 1, a high threshold Th_H = 30, and a low threshold Th_L = 10 was used.

  In step S93, the control device 1 calculates the average edge_M and the deviation edge_D of the edge strength of the character string candidate region. Based on the average edge_M and the deviation edge_D of the edge strength, the lower limit edge_L and the upper limit edge_H of the reference range of the edge strength are calculated using the following equations.

  Next, in step S94, the control device 1 calculates the average luminance I_M and deviation I_D of the character string candidate region. Based on the luminance average I_M and the deviation I_D, the lower limit I_L and the upper limit I_H of the luminance reference range are calculated using the following equations.

  In step S95, the control device 1 selects one character candidate in the selected character string candidate. In step S96, the control device 1 calculates the average edge strength of the selected character candidate region. In step S97, the control device 1 calculates the average brightness of the selected character candidate area. In step S98, when the selected character candidate has edge strength and brightness outside the reference range, the control device 1 deletes the character candidate. Specifically, a character candidate having an edge strength less than the lower limit edge_L or an edge strength greater than the upper limit edge_H is deleted as noise. In addition, character candidates having a luminance lower than the lower limit I_L or a luminance higher than the upper limit I_H are deleted as noise.

  Next, in step S99, the control device 1 determines whether or not there is an unprocessed character candidate. If YES, the process proceeds to step S95, and if NO, the process proceeds to step S100. In step S100, the control device 1 determines whether or not there is an unprocessed character string candidate. If YES, the process proceeds to step S90, and if NO, the process proceeds to step S69 in FIG.

  FIG. 40 is a diagram illustrating an example of an input image for explaining the edge strength and region luminance determination processing in step S68 of FIG. FIG. 41 is a diagram showing an example of character string candidates selected in step S90 of FIG. FIG. 42 is a diagram illustrating an example of character string candidates processed by the edge strength and region luminance determination processing in step S68 of FIG. 40 to 42, it can be seen that the noise is reduced based on the edge strength and the region luminance.

  In step S69 of FIG. 11, the control device 1 executes an average height determination process.

  FIG. 13 is a flowchart showing a subroutine of average height determination processing in step S69 of FIG. In step S101, the control device 1 selects one character string candidate. In step S102, the control device 1 connects the objects included in each character candidate in the height direction. In order to connect the objects in the height direction, a closing process is performed in the height direction (that is, an area expansion process is performed and then a contraction process is performed). In step S103, the control device 1 calculates an average height and deviation of each character candidate, and determines a reference range for the height. In order to determine the reference range of height, instead of calculating the average and the deviation, an intermediate value of the height of each object in the character string candidate may be calculated. In this case, for example, a range of 5 pixels or more and 1.1 times or less of the intermediate height value may be set as the reference range. In step S <b> 104, the control device 1 deletes a character candidate that does not have the height of the specified range, and extracts a new character string candidate separated by deleting the character candidate from the original character string candidate. FIG. 43A is a diagram illustrating an example of the character string candidates selected in step S101 of FIG. FIG. 43B is a diagram showing an example of a new character string candidate extracted in step S104 of FIG. The character candidate 42a is deleted as noise, and new character string candidates 41c and 41d are extracted. Next, in step S105 in FIG. 13, the control device 1 determines whether or not there is an unprocessed character string candidate. If YES, the process proceeds to step S101, and if NO, the process proceeds to step S70 in FIG.

  In step S70 of FIG. 11, the control device 1 determines whether or not the number of character string candidates is 0. If YES, the control device 1 proceeds to steps S52, S54, S56 of FIG. 9 or step S7 of FIG. If NO, the process proceeds to step S71. In step S71, the control device 1 selects one character string candidate. In step S72, the control device 1 executes date pattern determination processing.

  FIG. 14 is a flowchart showing a subroutine of date pattern determination processing in steps S72 and S75 of FIG.

  The control device 1 stores therein a table including a plurality of date patterns including a two-digit or four-digit number representing a year, a one-digit or two-digit number representing a month, and a predetermined punctuation mark. ing. Taking “July 2012” as an example, the date has the following pattern, for example.

[Table 1]
―――――――――――――――――――――――――――――――――
(1) “2012.07” 4 digits + period + 2 digits (2) “20122.7” 4 digits + period + 1 digit (3) “2012-07” 4 digits + hyphen + number 2 Digit (4) “2012-7” 4 digits + hyphen + 1 digit (5) “'12 .07” Apostrophe + 2 digits + period + 2 digits (6) “'12 .7” Apostrophe + numbers 2 digits + period + 1 digit (7) "201207" Precautions when period (1) is dirty (8) "2012, 7" Precautions when period (2) is dirty ―――――――――――――――――――――――――――――――――

  Each date pattern specifies how numbers and punctuation are arranged.

  In step S111 of FIG. 14, the control device 1 performs OCR for alphanumeric characters on the character string candidate. In step S112, the control device 1 selects one date pattern from the date patterns held in the internal table. In the following steps, the control device 1 determines whether or not the character string recognized in step S111 matches the date pattern selected in step S112. In step S113, the control device 1 determines whether or not the character string matches the date pattern. If YES, the process proceeds to step S114, and if NO, the process proceeds to step S115. In step S114, the control device 1 determines whether or not the character string includes “alphabetic characters that are clearly not mistaken for numbers”. If YES, the process proceeds to step S115. If NO, the process proceeds to step S117. “A letter that is clearly not mistaken for a number” is, for example, “O”, “o”, “C”, “c”, “U”, “u”, “Z”, “z”, “n”. , “L”, “l”, “I”, “J”, “D”. When the character string includes “alphabetic characters that are clearly not mistaken for numbers”, the character string is recognized as not a date. In step S115, the control device 1 determines whether or not all date patterns have been used. If YES, the control device 1 proceeds to step S73 (or step S76) in FIG. 11, and if NO, the process returns to step S112. Select a date pattern. In step S117, the control device 1 determines whether or not the height of the character in the character string is constant. If YES, the process proceeds to step S116, and if NO, the process returns to step S113. In step S116, the control device 1 determines whether or not the character string includes alphabetic characters. If YES, the process proceeds to step S117, and if NO, the process proceeds to step S118. In step S117, the control device 1 performs OCR for numbers on the character string candidate, and recognizes that the character string candidate includes an alphabetic character as a number.

  After step S117, tilt correction may be performed. By performing the inclination correction, a character string including “1” can be recognized without error.

  In step S118 of FIG. 14, the control device 1 determines whether or not the character string is a date. If YES, the process proceeds to step S73 (or step S76) of FIG. 11, and if NO, the process returns to step S113. .

  In step S73 in FIG. 11, the control device 1 determines whether or not the date pattern has been successfully determined. If YES, the process proceeds to steps S52, S54, S56 in FIG. 9 or step S7 in FIG. If YES, go to step S74. In step S74, the control device 1 rotates the character string candidate by 180 degrees. In step S75, the control device 1 executes the same date pattern determination process as described above for the character string candidate rotated by 180 degrees. In step S76, the control device 1 determines whether or not the date pattern has been successfully determined. If YES, the control device 1 proceeds to steps S52, S54, S56 in FIG. 9 or step S7 in FIG. 6, and if NO, Proceed to step S77. In step S77, the control device 1 determines whether or not there is an unprocessed character string candidate. If YES, the process returns to step S71. If NO, step S52, S54, S56 in FIG. 9 or FIG. The process proceeds to step S7.

  In step S7 of FIG. 6, the control device 1 determines whether or not the OCR has succeeded, that is, whether or not the character string representing the expiration date has been successfully extracted. If YES, the process proceeds to step S8. If NO, the process proceeds to step S10. When the month is recognized as “1”, it is actually “10” to “12”, but may be erroneously recognized as “1” due to the angle of the container 13 or the like. is there. In the following steps, when a character string candidate related to one input image includes only “1” as a number representing the month, does a character string candidate related to another input image include only “1” as a number representing the month? Judge whether or not. In step S8, the control device 1 determines whether or not the month is “1”. If YES, the process proceeds to step S9, and if NO, the process proceeds to step S12. In step S9, the control device 1 determines whether or not the same date has been detected twice. If YES, the process proceeds to step S12. If NO, the process proceeds to step S10. In step S10, the control device 1 determines whether or not the container 13 has made one turn, the process proceeds to step S13 if YES, and the process proceeds to step S11 if NO. In step S <b> 11, the control device 1 rotates the container 13. For example, when the container 13 is rotated by 15 degrees, a total of 24 input images can be acquired. In addition, even when the diameter of the container 13 is different, the diameter of the container 13 is detected by rotating the container 13 over two rounds in order to acquire an image at every fixed angle, and the image of the container 13 is acquired at a constant time interval However, the container 13 may be rotated at different speeds depending on the diameter. In step S12, the control device 1 outputs a date. In step S13, the control device 1 outputs an error.

  As described above, according to the optical character recognition device according to the present embodiment, various noises included in the image of the character string are removed in advance before optically recognizing the character string. A character string representing a date can be recognized with high accuracy.

  The input image is not limited to an image of a cylindrical container, and may be another image (a flat object image or arbitrary image data).

  When the control device 1 is connected to the external PC 9, the date detection process of FIGS. 6 to 14 may be executed at least partially by the PC 9.

  As described above, an optical character recognition method for recognizing a character string representing a date may be performed. Such an optical character recognition method may be implemented as a computer program that optically recognizes a character string when executed by a computer. Such a computer program may be stored in a computer-readable recording medium. For example, such a computer program is stored in the recording medium 10 of FIG. 1, and when the PC 9 reads the computer program from the recording medium 10, the optical character recognition method is performed according to the computer program.

Second embodiment.
The optical character recognition device according to the first embodiment recognizes a character string representing a date with higher accuracy than before if the date includes a predetermined punctuation and is printed with characters having a normal typeface. can do. However, dates that include special punctuation (for example, “2015/5”, “2015 5”), dates printed with characters that have special typefaces (for example, a plurality of dots separated from each other), etc. In order to recognize (hereinafter, referred to as a special pattern), it is necessary to extract a large number of character string candidates by relaxing various judgment conditions (threshold values, etc.) in the date detection processing of FIGS. is there. Relaxing the judgment conditions in the date detection process increases noise and increases the time required to execute the date detection process, so it is desirable to be able to recognize the date of the special pattern while suppressing an increase in execution time. .

  FIG. 44 is a flowchart showing date detection processing executed by the control device 1 of the optical character recognition device according to the second embodiment of the present invention. Steps S1 to S12 in FIG. 44 are the same as steps S1 to S12 in FIG. The date detection process in FIG. 44 includes a special pattern date detection process in step S14 instead of step S13 in FIG. In the date detection process of the special pattern in step S14, the judgment conditions (threshold value, etc.) that are more relaxed than those used in steps S2, S4, and S6 of FIG. 44 are set, and the date detection process of FIG. Run. There are few types and quantities of containers on which the date of the special pattern is printed, and the dates printed on most containers can be recognized by executing a date detection process in which limited conditions are set to some extent.

  According to the date detection process of FIG. 44, the date detection process of the special pattern of step S14 is executed only when the character string of the date is not recognized by executing steps S1 to S11 of FIG. The date of the special pattern can be recognized while suppressing the increase.

Third embodiment.
According to the optical character recognition device according to the first embodiment, since a plurality of images (input images) each representing different angles of the container 13 are acquired, the date character string is orthogonal to the rotation axis of the cylindrical container. (FIGS. 3 and 4), there is a possibility that the entire date does not fit in the image. When the date string extends over a half circumference of the side surface of the cylindrical container, an image including the entire date cannot be acquired.

  FIG. 45 is a flowchart showing date detection processing executed by the control device 1 of the optical character recognition device according to the third embodiment of the present invention. Steps S1 to S12 in FIG. 45 are the same as steps S1 to S12 in FIG. The date detection process in FIG. 45 includes a date detection process for linked images in step S15 instead of step S13 in FIG. FIG. 46 is a flowchart showing a subroutine of date detection processing for linked images in step S15 of FIG. As described above, the control device 1 photographs the container 13 with the camera 5 while rotating the container 13 by a certain angle using the rollers 8a and 8b, and acquires a plurality of images representing different angles of the container 13, respectively. . In step S1A of FIG. 46, the control device 1 connects a plurality of images including portions adjacent to each other of the container 13 to generate one connected image. Specifically, the control device 1 connects these images by recognizing a similar object in two adjacent images. The connected image is a planar image in which the side surface of the container is developed. The control device 1 corrects the curved portion of the container 13 to a plane using projective transformation, with the width of the container 13 as the diameter of the cylinder. Steps S2 to S9, S12, and S13 in FIG. 46 are the same as steps S2 to S9, S12, and S13 in FIG.

  Since it takes a certain amount of time to generate a connected image, the control device 1 may generate a connected image in advance before executing step S15 of FIG.

  Conventionally, a line camera is known for generating an image of a side surface of a cylindrical object. However, in order to use a line camera, in addition to the cost of the line camera itself, there is a cost to provide a mechanism for rotating an object with high accuracy. Since the medicine container has various shapes and sizes, the cost for rotating the medicine container with sufficient accuracy for photographing with a line camera becomes very large. On the other hand, according to the date detection process of FIG. 45, it is possible to suppress an increase in cost by connecting a plurality of images captured by a normal camera to generate a connected image.

  According to the date detection process of FIG. 45, the date detection process of the linked image in step S15 is executed only when the character string of the date cannot be recognized by executing steps S1 to S11 of FIG. Dates that do not fall within one image can be recognized while suppressing the increase.

Fourth embodiment.
When there are other characters ("O", "J", "Z", etc.) immediately after (right side) the date string, there is a possibility that other characters may be mistakenly recognized as part of the date. is there. Therefore, it is necessary to remove other characters that are not part of the date from the character string candidates.

  FIG. 47 is a flowchart showing a subroutine of date pattern determination processing of date detection processing executed by the control device 1 of the optical character recognition device according to the fourth embodiment of the present invention. The date pattern determination process in FIG. 47 is executed in steps S72 and S75 in FIG. 11, and includes additional steps S121 and S122 between steps S113 and S114 in FIG. In step S121 of FIG. 47, when the number representing the month of the character string matching the date pattern is any one of “10”, “11”, and “12”, the control device 1 is based on the criteria described below. Whether or not there are other characters after the date is determined. If YES, the process proceeds to step S122. If NO, the process proceeds to step S114.

  FIG. 48 is a diagram illustrating an example of a character string candidate including a date character string and other characters. Another character “CJ932” exists after the date “20166.1”. If “C” is mistakenly recognized as “0”, the character string matching the date pattern is mistakenly recognized as “2016.10”. When detecting the date character string “2016.10”, the control device 1 determines whether or not the last “0” is actually a part of the date.

  The distances D1 to D10 between characters in FIG. 48 are, for example, as follows, with the number of pixels as a unit.

[Table 2]
――――――――――――――
D1 = 50.4231
D2 = 49.8108
D3 = 43.7262
D4 = 44.1873
D5 = 82.874
D6 = 82.8974
D7 = 55.6709
D8 = 39.335
D9 = 49.1936
D10 = 45.8404
――――――――――――――

  In order to determine whether or not the last “0” of “2016.10” is a part of the date, first, consider a case where the distances D1 to D6 between the date characters are compared. In this case, the distance between the last “1” and “0” (the distance between “1” and “C”) D6 is approximately the same as the distance D5 between “.” And “1”. .10 ”cannot be determined as the last“ 0 ”is not part of the date.

  In the present embodiment, in order to determine whether or not the last “0” of “2016.10” is part of the date, the distance D6 and the distance D7 between characters after “2016.10.” The average value of D10 is compared. When the distance D6 is larger than the average value of the distances D7 to D10, the control device 1 determines that the last “0” of “2016.10” is not part of the date, and in step S122, the character “CJ932” is determined. Remove. On the other hand, when the distance D6 is equal to or less than the average value of the distances D7 to D10, the control device 1 determines that the last “0” of “2016.10” is a part of the date. Thereby, even if there is another character immediately after the date character string, the character string representing the date can be recognized with high accuracy.

  As described above, the control device 1 uses the character string candidate when the character string candidate includes two numbers representing the month and at least one other character following the two numbers representing the month, and 2 representing the month. When the distance between two numbers is greater than the average of the distance between the two numbers representing the month and other characters and the distance between the other characters (step S121), the first digit of the two numbers representing the month And other characters are removed (step S122).

  The optical character recognition device, the optical character recognition method, the computer program, and the recording medium of the present invention can recognize a character string representing a date with higher accuracy than before.

1 ... Control device,
2 ... Rail,
3 ... mobile device,
4-6 ... Camera,
7a, 7b ... lighting device,
8a, 8b ... rollers,
9 ... Personal computer (PC),
10: Recording medium,
11,12 ... Tray,
13, 13a-13d ... container,
21 ... Target area,
22 ... Trimmed target area,
31 ... Character string candidate mask,
41 ... The bounding box of the character string candidate,
41a, 41b ... bounding boxes for new character string candidates,
42, 42a ... A bounding box of character candidates,
43 ... A bounding box of transformed character candidates,
51 ... Character candidate.

Claims (17)

In an optical character recognition device that optically recognizes a character string, the optical character recognition device includes:
A first processing means for extracting a target area including an object to be recognized from an input image including an image of a container or an image of a label attached to the container ;
Second processing means for extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. Third processing means for recognizing column candidates as dates ;
An optical character recognition device comprising: a fourth processing unit for determining whether or not to store the container in at least one storage, based on the date recognized by the third processing unit. . The second processing means includes
Detect the contour and edge of the object included in the target area,
2. The optical character recognition apparatus according to claim 1, wherein an object having an outline and an edge that overlap each other is extracted as the candidate object. The second processing means includes
Applying a Sobel filter to the target area to detect the first edge;
Applying a canny filter to a region in the vicinity of the first edge to detect a second edge;
The optical character recognition apparatus according to claim 2, wherein the second edge is used as an edge of an object included in the target area. The third processing means includes
Label the above character string candidate objects to extract multiple character candidates,
A plurality of bounding boxes each having a rectangular shape having a width parallel to a direction in which the character string candidates extend and a height orthogonal to the direction in which the character string candidates extend, and surrounding each of the character candidates. Generate and
Each of the bounding boxes is deformed so that the width of the bounding box is increased as the height of the bounding box is lower.
The optical character recognition apparatus according to claim 1, wherein a set of character candidates included in bounding boxes connected by deformation is extracted as a new character string candidate. The third processing means includes
Label the above character string candidate objects to extract multiple character candidates,
The optical character recognition apparatus according to claim 1, wherein a character string candidate including more than ten character candidates is deleted. The third processing means includes
Label the above character string candidate objects to extract multiple character candidates,
The character string candidate including only a character candidate including two or more objects in a direction orthogonal to a direction in which the character string candidate extends is deleted. The optical character recognition device described. The third processing means includes
Detect the contours and edges of the above character string candidate objects,
7. The character string candidate in which a portion where the edge pixel coincides with the contour pixel is 60% or less of the area of the edge pixel is deleted. The optical character recognition device described.   The third processing means recognizes the character string candidate as not being a date when the character string candidate includes an alphabetic character that is clearly not mistaken for a number. The optical character recognition apparatus as described in any one of these.   When the character string candidate includes two numbers representing the month and at least one other character subsequent to the two numbers representing the month, the third processing unit is configured to represent the month. When the distance between two numbers is larger than the average of the distance between the two numbers representing the month and other characters and the average distance between the other characters, the first digit of the two numbers representing the month and The optical character recognition apparatus according to claim 1, wherein the other character is removed.   The optical character recognition device according to claim 1, wherein the input image is an image of a cylindrical container that is rotatably held.   The first processing means includes, from the input image, an edge extending in a direction substantially perpendicular to a rotation axis of the cylindrical container, and a portion having a luminance higher than a predetermined threshold value. The optical character recognition apparatus according to claim 10, wherein a region to be included is extracted as the target region. The optical character recognition device acquires a plurality of input images respectively representing different angles of the container photographed while rotating the container,
When the character string candidate related to one input image includes only “1” as the number representing the month, the third processing means only includes “1” as the number representing the month as the character string candidate related to the other input image. The optical character recognition device according to claim 10 or 11, wherein it is determined whether or not the character string is included. The optical character recognition device is
A camera,
An imaging table for holding the container so as to be rotatable around a rotation axis of the cylindrical container;
A moving device for moving the container between at least one storage and the imaging table;
The optical character recognition device according to any one of claims 10 to 12, wherein a character string representing a date of expiration date of the medicine in the container is printed on the container. In the optical character recognition method for optically recognizing a character string, the optical character recognition method includes:
A first step of extracting a target area including an object to be recognized from an input image including an image of a container or an image of a label attached to the container ;
A second step of extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. A third step of recognizing column candidates as dates ;
An optical character recognition method comprising: a fourth step of determining whether to store the container in at least one storage, based on the date recognized in the third step . In a computer program that optically recognizes a character string when executed by a computer, the computer program includes:
A first step of extracting a target area including an object to be recognized from an input image including an image of a container or an image of a label attached to the container ;
A second step of extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. A third step of recognizing column candidates as dates ;
A computer program comprising: a fourth step of determining whether to store the container in at least one storage, based on the date recognized in the third step . In a computer-readable recording medium storing a computer program for optically recognizing a character string when executed by a computer, the computer program includes:
A first step of extracting a target area including an object to be recognized from an input image including an image of a container or an image of a label attached to the container ;
A second step of extracting candidate objects including at least one character string candidate object from the objects included in the target area;
By labeling the candidate objects, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and the character string candidates are two-digit or four-digit numbers representing the year. And whether or not the character string candidate has the date pattern, the character string has a date pattern that includes a one- or two-digit number representing the month and a predetermined punctuation mark. A third step of recognizing column candidates as dates ;
A recording medium comprising: a fourth step of determining whether to store the container in at least one storage, based on the date recognized in the third step . In an optical character recognition device that optically recognizes a character string, the optical character recognition device includes:
First processing means for extracting a target area including an object to be recognized from an input image;
Second processing means for extracting candidate objects including at least one character string candidate object from the objects included in the target area;
When the candidate objects are labeled, a plurality of objects extending in a predetermined direction and close to each other are extracted as the character string candidates, and when the character string candidates have a predetermined date pattern, Third processing means for recognizing the character string candidate as a date,
The input image is an image of a cylindrical container held rotatably,
The first processing means includes, from the input image, an edge extending in a direction substantially perpendicular to a rotation axis of the cylindrical container, and a portion having a luminance higher than a predetermined threshold value. An optical character recognition apparatus, wherein a region including the target region is extracted as the target region.

Images