TWI608422B - Optical character recognition device, optical character recognition method, and recording medium - Google Patents

Description

Optical character recognition device, optical character recognition method and recording medium

The present invention relates to an optical character recognition apparatus for optically recognizing a character string, and more particularly to an optical character recognition apparatus for recognizing a character string representing a date. Further, the present invention relates to an optical character recognition method, a computer program, and a recording medium for identifying a character string representing a date.

There is a need for an apparatus for optically identifying a character printed on a container of a medicine (refer to Patent Document 1). For example, when a drug such as an injection that has not been used for temporary delivery to a ward is returned to the storage, the medicine must be sorted according to the type, name, and expiration date of the medicine in order to promptly and unambiguously take out the medicine for the next use. And keep it. When an optical character recognition device is used and a return device that automatically performs the classification is realized, it is effective for work efficiency and reduction of errors. Moreover, since the storage place of the medicine is recorded when the medicine is stored by using such a returning device, the appropriate medicine can be automatically taken out according to the prescription at the time of the next use.

[Previous Technical Literature] [Patent Literature]

[Patent Document 1] Japanese Patent No. 4857768.

When the character string is optically recognized by the optical character recognition device, there is a possibility that erroneous recognition occurs due to various noises contained in the image of the character string. To improve The accuracy of the recognition must be removed from the image in advance.

The present invention solves the above problems, and provides an optical character recognition device, an optical character recognition method, a computer program, and a recording medium capable of recognizing a character string indicating a date higher than the previous precision.

According to the optical character recognition device of the first aspect of the present invention, the optical character recognition device is characterized in that the optical character recognition device includes a first processing unit that extracts an object including the recognition target from the input image. a target region; a second processing unit that extracts a candidate object including at least one character string candidate from the object included in the target region; and a third processing unit that performs the candidate object Marking, and extracting a plurality of objects extending in a predetermined direction and adjacent to each other as the character string candidate, and determining whether the character string candidate has a 2-digit or 4-digit number indicating a year, indicating 1 or 2 digits of the month The pattern of the number and the date of the predetermined punctuation mark recognizes the character string candidate as a date when the character string candidate has the pattern of the date.

An optical character recognition device according to a second aspect of the present invention is the optical character recognition device according to the first aspect, characterized in that the second processing means detects an outline of an object included in the target region. And an edge, and an object having contours and edges overlapping each other is extracted as the candidate object.

An optical character recognition device according to a third aspect of the present invention is the optical character recognition device according to the second aspect, characterized in that the second processing means is: applying a Sobel filter to the target region (Sobel Filter) ) and detect the first side The edge is detected by applying a Canny filter to a region in the vicinity of the first edge, and the second edge is used as an edge of the object included in the target region.

An optical character recognition device according to a fourth aspect of the present invention, characterized in that the third processing means is configured to perform the character string candidate Marking the object and extracting a plurality of character candidates, generating a plurality of bounding boxes, wherein the plurality of bounding boxes respectively have a width parallel to a direction in which the character string candidates extend and a height orthogonal to a direction in which the character string candidates extend And each of the bounding frames is deformed so as to increase the width of the bounding frame as the height of the bounding frame is lower, and the bounding box included in the bounding box connected by the deformation is extracted. The group of text candidates is used as a new string candidate.

An optical character recognition device according to a fifth aspect of the present invention, characterized in that the third processing means is configured to perform the character string candidate. Marking the object and extracting a plurality of character candidates, and deleting the character string candidates including more than 10 character candidates.

An optical character recognition device according to a sixth aspect of the present invention, characterized in that the third processing means is configured to perform the character string candidate. The object object is marked and a plurality of character candidates are extracted, and only the character string candidates including the character candidates including two or more objects in the direction orthogonal to the direction in which the character string candidates extend are deleted.

An optical character recognition device according to a seventh aspect of the present invention, characterized in that the third processing means is configured to detect the character string candidate The outline and the edge of the object are deleted from the pixel of the edge and the pixel of the outline is a character string candidate of 60% or less of the area of the pixel of the edge.

An optical character recognition device according to an eighth aspect of the present invention, characterized in that the third processing means includes the string candidate When the English character of the number is not mistaken, the above character string candidate is recognized as a non-date.

An optical character recognition device according to a ninth aspect of the present invention, characterized in that the third processing means includes the character string candidate 2 digits of the month, and at least one other character following the 2 digits of the month, and the distance between the 2 digits of the month is greater than the distance between the 2 digits of the month and other characters When the average value of the distance between the other characters is the same, the one-digit number indicating the two digits of the month and the other characters are removed.

An optical character recognition device according to a tenth aspect of the present invention, characterized in that the input image is a cylinder that is rotatably held An image of a shape container.

An optical character recognition device according to an eleventh aspect of the present invention, characterized in that the first processing means extracts the input image from the input image substantially An edge extending in a direction in which the rotation axis of the cylindrical container is orthogonal to each other and a region having a luminance higher than a predetermined threshold value are used as the target region.

An optical character recognition device according to a twelfth aspect of the present invention is the tenth or eleventh An optical character recognition device according to the aspect of the invention, characterized in that the optical character recognition device is configured to obtain a plurality of input images respectively indicating different angles of the container while rotating the container, and the third processing mechanism is in one input image When the character string candidate contains only "1" as the number indicating the month, it is determined whether or not the character string candidate of the other input image contains only "1" as the number indicating the month.

According to a thirteenth aspect of the present invention, there is provided an optical character recognition device according to any one of the tenth to twelfth aspects, wherein the optical character recognition device comprises: a camera; a photographing station; Holding the container so as to be rotatable about a rotation axis of the cylindrical container; and moving the device to move the container between at least one of the storage and the photographing table; and printing on the container A string of the date of use of the drug in the above container.

According to a fourteenth aspect of the present invention, an optical character recognition method for optically recognizing a character string, wherein the optical character recognition method includes the first step of extracting an object including an identification object from an input image. a target region; a second step of extracting a candidate object including at least one character string candidate from the object included in the target region; and a third step of performing the candidate object Marking and extracting a plurality of objects extending in a predetermined direction and adjacent to each other as the character string candidate, and determining whether the character string candidate has a 2-digit or 4-digit number indicating the year, and indicates 1 or 2 of the month. The pattern of the digits and the date of the predetermined punctuation mark recognizes the character string candidate as a date when the character string candidate has the pattern of the date.

A computer program according to a fifteenth aspect of the present invention, which is a computer program for optically recognizing a character string when executed by a computer, the computer program comprising: the first step of extracting from an input image a target region including an object to be identified; and a second step of extracting a candidate object including at least one character string candidate from the object included in the target region; and a third step Marking the candidate object, extracting a plurality of objects extending in a predetermined direction and adjacent to each other as the character string candidate, and determining whether the character string candidate includes a 2-digit or 4-digit number indicating the year, indicating the month The one or two digits and the pattern of the date of the predetermined punctuation mark recognize the character string candidate as the date when the character string candidate has the pattern of the date.

A recording medium according to a sixteenth aspect of the present invention, which is a computer-readable recording medium storing a computer program for optically recognizing a character string when executed by a computer, wherein the computer program comprises: a first step of extracting a target region including an object to be recognized from the input image; and a second step of extracting an candidate including at least one character string candidate from the object included in the target region And a third step of performing marking of the candidate object, extracting a plurality of objects extending in a predetermined direction and adjacent to each other as the character string candidate, and determining whether the character string candidate has a representation year The pattern of the two or four digits, the one or two digits of the month, and the date of the predetermined punctuation mark, when the character string candidate has the pattern of the date, recognizes the character string candidate as the date.

An optical character recognition device according to a seventeenth aspect of the present invention, wherein the optical character recognition device is characterized in that: the optical character recognition device is configured to take a side of the container while rotating A plurality of input images representing different angles of the containers are respectively connected to 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 indicating a date higher than the previous precision.

1‧‧‧Control device

2‧‧‧ Track

3‧‧‧Mobile devices

4~6‧‧‧ camera

7a, 7b‧‧‧ lighting fixtures

8a, 8b‧‧‧roller

9‧‧‧Personal Computer (PC)

10‧‧‧Recording media

11, 12‧‧‧ tray

13, 13a~13d‧‧‧ containers

21‧‧‧ Target area

22‧‧‧Fixed target area

31‧‧‧String candidate mask

41‧‧‧Boundary bounds for string candidates

41a, 41b‧‧‧ new string candidate bounding box

42. 42a‧‧‧ bounds box

43‧‧‧The bounding box of the text of the variant

51‧‧‧ text alternate

D1~D10‧‧‧Distance

H1, h2‧‧‧ height

S1~S118‧‧‧Steps

W1, w2‧‧‧ width

Fig. 1 is a block diagram showing the configuration of an optical character recognition device according to a first embodiment of the present invention.

Fig. 2 is a plan view showing a container 13a in which a character string is printed in the first example.

Fig. 3 is a plan view showing a container 13b in which a character string is printed in the second example.

Fig. 4 is a plan view showing a container 13c in which a character string is printed in the third example.

Fig. 5 is a plan view showing a container 13d in which a character string is printed in the fourth example.

Fig. 6 is a flow chart showing the date detecting process executed by the control device 1 of Fig. 1.

Fig. 7 is a flow chart showing the subroutine of the target region extraction processing in step S2 of Fig. 6.

FIG. 8 is a flowchart showing a subroutine of the candidate object extraction processing in step S4 of FIG. 6.

Fig. 9 is a flow chart showing the subroutine of the OCR processing in step S6 of Fig. 6.

Fig. 10 is a flow chart showing the first part of the OCR subroutine in steps S51, S53, S55, and S57 of Fig. 9.

Fig. 11 is a flow chart showing a second part of the OCR subroutine in steps S51, S53, S55, and S57 of Fig. 9.

Figure 12 is a diagram showing the edge intensity and area brightness determination processing in step S68 of Figure 11 The subroutine flow chart.

Fig. 13 is a flow chart showing the subroutine of the average height determination processing in step S69 of Fig. 11.

Fig. 14 is a flowchart showing the subroutine of the date pattern determination processing in steps S72 and S75 of Fig. 11.

Fig. 15 is a view showing an example of a portion where the luminance in the image extracted in step S21 of Fig. 7 is high.

Fig. 16 is a view showing an example of long longitudinal edges in the image extracted in step S22 of Fig. 7.

Fig. 17 is a view in which the portion having the higher luminance of Fig. 15 and the longer longitudinal edge of Fig. 16 are overlapped.

Fig. 18 is a view showing a target region 21 including a portion having a higher luminance of Fig. 15 and a longer longitudinal edge of Fig. 16.

Fig. 19 is a view showing an example of a bright object extracted in step S31 of Fig. 8.

Fig. 20 is a view showing an example of a dark object extracted in step S33 of Fig. 8.

Fig. 21 is a view showing the extraction of the bright object using the moving average filtering method in step S31 of Fig. 8.

Fig. 22A is a view showing an image including a bright object and a dark object.

Fig. 22B is a view showing the extraction of the object from the image of Fig. 22A.

Fig. 22C is a view showing the extraction of a dark object from the image of Fig. 22A.

Fig. 23 is a view showing an example of edges in an image extracted by the Sobel filtering method in step S35 of Fig. 8.

Fig. 24 is a view showing an example of an image obtained by deleting an area other than the edge extracted in step S35 in step S36 of Fig. 8.

Fig. 25 is a view showing an example of edges in an image extracted using the Canni filter method of the threshold value 15 in step S37 of Fig. 8.

Fig. 26 is a view showing an example of an edge in an image extracted by the Canni filter method of the threshold 4 in step S37 of Fig. 8.

FIG. 27 is a view showing an example of a candidate object of the bright object extracted in step S38 of FIG. 8.

FIG. 28 is a view showing an example of a candidate object of a dark object extracted in step S39 of FIG. 8.

Fig. 29A is a view showing an example of an object.

FIG. 29B is a diagram showing the binarized image of FIG. 29A extracted using the threshold 200.

Fig. 29C is a view showing the outline of Fig. 29B.

Figure 29D is a diagram showing the edge of Figure 29A extracted using threshold 50.

Figure 29E is a diagram showing the edge of Figure 29A extracted using the threshold 200.

Figure 29F is a diagram showing the outline and edges of a character.

Fig. 30 is a view showing an example of character string candidates extracted in step S61 of Fig. 10;

Fig. 31 is a view showing the extraction of character string candidates in step S61 of Fig. 10.

Fig. 32 is a view showing an example of a character candidate extracted in step S62 of Fig. 10 and a boundary block generated.

Fig. 33A is a view showing an example of character string candidates extracted in step S61 of Fig. 10;

Fig. 33B is a view showing an example of the character candidates extracted in step S62 of Fig. 10 and the resulting bounding block 42.

Fig. 33C is a view showing an example of the bounding frame 43 of the deformation in the step S63 of Fig. 10.

Fig. 33D is a diagram showing an example of a new character string candidate extracted in step S64 of Fig. 10.

Fig. 34A is a view showing another example of the character string candidates extracted in step S61 of Fig. 10;

Fig. 34B is a view showing another example of the character candidates extracted in step S62 of Fig. 10 and the resulting bounding block 42.

Figure 35 is a diagram showing the deletion of a part of the string in steps S65, S66, and S67 of Figure 10 A diagram of an example of a candidate for a candidate after the candidate.

Fig. 36A is a view showing an example of character string candidates extracted in step S64 of Fig. 10;

FIG. 36B is a view showing the number of objects in the height direction of each character candidate included in the character string candidate of FIG. 36A.

Fig. 37A is a view showing another example of the character string candidates extracted in step S64 of Fig. 10;

37B is a view showing the number of objects in the height direction of each character candidate included in the character string candidate of FIG. 37A.

Fig. 38A is a view showing an example of an input image.

Fig. 38B is a view showing a candidate object of the bright object extracted from the image of Fig. 38A.

Fig. 38C is a view showing the outline of the candidate object of Fig. 38B.

Figure 38D is a diagram showing the edges extracted from the image of Figure 38A.

Fig. 39A is a view showing an example of an input image.

Fig. 39B is a view showing a candidate object of a dark object extracted from the image of Fig. 39A.

Fig. 39C is a view showing the outline of the candidate object of Fig. 39B.

Figure 39D is a diagram showing the edge extracted from the image of Figure 39A.

Fig. 40 is a view showing an example of an input image for explaining the edge intensity and the area brightness determination processing in step S68 of Fig. 11.

Fig. 41 is a view showing an example of character string candidates selected in step S90 of Fig. 12;

Fig. 42 is a view showing an example of character string candidates processed by the edge intensity and region luminance determination processing in step S68 of Fig. 11;

Fig. 43A is a view showing an example of character string candidates selected in step S101 of Fig. 13;

Fig. 43B is a view showing an example of a new character string candidate extracted in step S104 of Fig. 13;

Fig. 44 is a flow chart showing the date detection processing executed by the control device 1 of the optical character recognition device according to the second embodiment of the present invention.

Fig. 45 is a flowchart showing the date detection processing executed by the control device 1 of the optical character recognition device according to the third embodiment of the present invention.

Fig. 46 is a flow chart showing the subroutine of the date detection processing of the concatenated image in step S15 of Fig. 45.

FIG. 47 is a flowchart showing a subroutine of the date pattern determination processing of the date detection processing executed by the control device 1 of the optical character recognition device according to the fourth embodiment of the present invention.

Fig. 48 is a view showing an example of a character string candidate including a character string of a date and other characters.

The first embodiment.

Fig. 1 is a block diagram showing the configuration of an optical character recognition device according to a first embodiment of the present invention. The optical character recognition device of Fig. 1 is a character string optically identifying the date printed on the surface of the cylindrical container 13.

The optical character recognition device of Fig. 1 includes a control device 1, a track 2, a moving device 3, cameras 4-6, illumination devices 7a, 7b, and rollers 8a, 8b. At least two rollers 8a, 8b are disposed in parallel with each other and include a driving device that operates under the control of the control device 1, whereby the container 13 is rotatably held. The optical character recognition device further includes at least one tray (or storage) 11, 12 that houses the container 13. The mobile device 3 is moved under the control of the control device 1 to move the container 13 between the trays 11, 12 and the rollers 8a, 8b. The cameras 4 to 6 are respectively disposed on the trays 11 and 12 and the rollers 8a and 8b. When the containers 13 are placed on the trays 11 and 12 and the rollers 8a and 8b, the images of the containers 13 are respectively taken and transmitted to the control device 1. The illuminating devices 7a, 7b illuminate the containers 13 on the drums 8a, 8b. The rollers 8a and 8b and the illuminating devices 7a and 7b function as a photographic table for the container 13. Optical character recognition device can also Instead of the rollers 8a, 8b, other mechanisms for holding the container 13 in a manner rotatable about the axis of rotation of the cylindrical container 13 are included. The control device 1 performs the following date detection processing on the image of the container 13 transported from the camera 5 with reference to Figs. 6 to 14 and recognizes the date printed on the surface of the container 13. An additional camera can also be provided on the mobile device 3. The control device 1 can also be connected to a personal computer (PC) 9 that operates in accordance with a computer program read from the recording medium 10.

The container 13 is, for example, a container (ampoules) for medicines, and a character string indicating the date of use 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 uses the moving device 3 to move the container 13 from the tray 11 to the rollers 8a, 8b, and optically recognizes the rollers 8a, 8b. The date of expiration date printed on the container 13. Then, the optical character recognition device determines whether to store the container or the waste container based on the date of identification, and uses the moving device 3 to move the container 13 to an appropriate storage or other tray 12 associated with the garbage can.

2 to 5 show an example of a character string printed on the container 13. Fig. 2 is a plan view showing a container 13a in which a character string is printed in the first example. Fig. 3 is a plan view showing a container 13b in which a character string is printed in the second example. Fig. 4 is a plan view showing a container 13c in which a character string is printed in the third example. Fig. 5 is a plan view showing a container 13d in which a character string is printed in the fourth example. The string can be printed either on the label attached to the container or directly on the container. Further, the orientation of the character string may be parallel to the rotation axis of the cylindrical container 13, or may be orthogonal to the rotation axis of the container 13, or may be mixed with the orientation strings.

The control device 1 operates as a first processing unit that extracts a target region including an object to be recognized from an input image. In addition, the control device 1 operates as a second processing unit that extracts a candidate object including at least one character string candidate from the object included in the target region. Further, the control device 1 operates as a third processing unit that marks the candidate object and extracts a plurality of objects extending in a predetermined direction and adjacent to each other as character string candidates, and determines the character string. Whether the candidate has an inclusion A pattern indicating the date of 2 or 4 digits of the year, the 1st or 2 digits of the month, and the date of the pre-defined punctuation mark. When the character string candidate has a date pattern, the character string candidate is recognized as date.

Hereinafter, the date detection processing executed by the control device 1 of Fig. 1 will be described with reference to Figs. 6 to 14 .

Fig. 6 is a flow chart showing the date detecting process executed by the control device 1 of Fig. 1. The control device 1 uses the rollers 8a and 8b to rotate the container 13 by a fixed angle (for example, 15 degrees) each time, and the container 13 photographs the container 13 to obtain a plurality of images respectively indicating different angles of the container 13 ( Enter the image). As the camera 5, a person having sufficient resolution is used to optically recognize the character string printed on the container 13. For example, the container 13 has a diameter of 10 to 40 mm, and for example, a black-and-white camera that photographs a range of 120 × 90 mm including the container 13 in a pixel number of 3840 × 2748 (about 10 million pixels) can be used. In this case, 1 mm on the container 13 corresponds to 32 pixels. In step S1 of FIG. 6, the control device 1 acquires one of a plurality of images of the container 13. In step S2, the control device 1 performs target region extraction processing.

Fig. 7 is a flow chart showing the subroutine of the target region extraction processing in step S2 of Fig. 6.

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 reflection of illumination) from the image acquired in step S1. The control device 1 extracts, for example, a portion having a luminance higher than 220 when the luminance of the pixel is varied in the range of 0 to 255. Fig. 15 is a view showing an example of a portion where the luminance in the image extracted in step S21 of Fig. 7 is high. Here, the input image is an image of the container 13d of FIG.

Then, 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. Although the rollers 8a, 8b are present in the background of the container 13, since the rollers 8a, 8b extend parallel to the axis of rotation of the container 13, they can be removed by extracting the longitudinal edges. Except for the influence of the rollers 8a, 8b.

In order to extract the edge in step S22, a Sobe1 filtering method of the following formula may be used.

The shorter of the edges extracted by the Sobel filtering method, for example, those that do not satisfy the length of 55 pixels are deleted as noise. Fig. 16 is a view showing an example of a longer vertical edge in the image extracted in step S22 of Fig. 7. The rotation axis of the container 13d of Fig. 5 is parallel to the X-axis of Fig. 5, so in step S22, an edge extending in a direction substantially parallel to the Y-axis of Fig. 5 is extracted.

In step S23 of Fig. 7, the control device 1 extracts a rectangular region (width w1 × height h1) including a portion having a high luminance and a vertical edge as a target region, and deletes an area outside the target region. Fig. 17 is a view in which the portion having the higher luminance of Fig. 15 and the longer longitudinal edge of Fig. 16 are overlapped. Fig. 18 is a view showing a target region 21 including a portion having a higher luminance of Fig. 15 and a longer longitudinal edge of Fig. 16. The target area is an area that is considered to contain an object that identifies a character string of the object.

Referring again to FIG. 6, after the target region extraction processing of step S2 is performed, in step S3, the control device 1 determines whether or not the extraction of the target region is successful. If YES, the process proceeds to step S4, and when 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 the candidate object extraction processing in step S4 of FIG. 6.

In step S31 of FIG. 8, the control device 1 applies a movement to the image of the target area. The average filtering method extracts objects that are brighter than the surrounding and binarizes the image. A string of white characters in a black background is extracted as a bright object. Since there is unevenness in the manner in which illumination is performed, it is not possible to detect an object by simply performing binarization. Therefore, a binarization method (dynamic threshold method) using a moving average filtering method is used. Fig. 21 is a view showing the extraction of the bright object using the moving average filtering method in step S31 of Fig. 8. According to the principle shown in Fig. 21, the control device 1 calculates the local average brightness based on the brightness of the input image (here, the image of the target area), and extracts the average brightness above the local plus the specific amount of deviation. The object of the brightness obtained by the brightness (that is, the object having the outstanding brightness compared with the surroundings) is used as the bright object. The size of the area to be referred to in order to calculate the local average brightness is determined according to the size of the entire target area. Fig. 19 is a view showing an example of a bright object extracted in step S31 of Fig. 8. Then, in step S32, the control device 1 detects the contour of the binarized object.

Then, in steps S33 to S34 of FIG. 8, the dark object is subjected to the same processing as the processing of the bright object in steps S31 to S32. In step S33, the control device 1 extracts a dark object darker than the surroundings by applying a moving average filtering method to the image of the target area, and binarizes the image. The string of black characters in the white background is extracted as a dark object. Fig. 20 is a view showing an example of a dark object extracted in step S33 of Fig. 8. In step S34, the control device 1 detects the contour of the binarized dark object.

Fig. 22A is a view showing an image including a bright object and a dark object. Fig. 22B is a view showing the extraction of the object from the image of Fig. 22A. Fig. 22C is a view showing the 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 of Fig. 8, the date printed on the container 13 can be surely detected.

Then, in step S35 of Fig. 8, the control device 1 extracts the edge in the image of the target area using the Sobel filtering method of the following formula.

Here, b is a result obtained by applying the operator B to a certain pixel, and c is a result obtained by applying the operator C to the same pixel. Fig. 23 is a view showing an example of edges in an image extracted by the Sobel filtering method in step S35 of Fig. 8.

Then, in step S36 of Fig. 8, the control device 1 deletes the area other than the edge extracted in step S35 from the image of the target area. Fig. 24 is a view showing an example of an image obtained by deleting an area other than the edge extracted in step S35 in step S36 of Fig. 8. Then, in step S37 of Fig. 8, the control device 1 applies the Canny filtering method to the image subjected to the deletion in step S36, and extracts the edges in the image.

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

Here, f _x (x, y) represents a product of a first-order numeracy and a pixel value function with respect to the x direction of the Gaussian function having the standard deviation σ, and f _y (x, y) represents the same Gaussian function. The first order of the y direction is a product of the product of the pixel and the value of the pixel value.

As the second step of the Canney edge detection method, the edge is detected by finding the maximum value of the magnitude g(x, y) of the slope. At this time, using the 8 pixels around the pixel of interest, the magnitude of the slope of the interpolation with respect to the slope d(x, y) is estimated, and the estimated values are compared, thereby determining the magnitude of the slope of the pixel of interest g(x, y) Is there a true maximum?

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

In the example of the present disclosure, in step S37, a Canny filtering method having a standard deviation σ=1.4, a high threshold Th_H=10, and a low threshold Th_L=5 of a Gaussian function is used. The range of threshold values is 0~255.

In step 40 of Figure 8, the two edges are extracted using two Canny filtering methods for use in both subsequent steps S38, S39, and step S67 of Figure 10 below. Figure 25 is a view showing the edge in the image extracted by the Canni filter method using the threshold of 15 in the step S37 of Figure 8. A diagram of the example of the edge. Fig. 26 is a view showing an example of an edge in an image extracted by the Canni filter method of the threshold 4 in step S37 of Fig. 8. In the example of the present disclosure, the edges of FIG. 26 are used in steps S38, S39, and the edges of FIG. 25 are used in step S67 of FIG.

If the edge is extracted using Sobel filtering, it is high speed, but the width of the extracted edge will be wider. If the edge is extracted using Canny filtering, it is low speed, but the detailed edge can be extracted. On the other hand, in the present embodiment, the edge is temporarily extracted using the Sobel filter method (step S35), and the area other than the extracted edge is deleted (step S36), and only the Caney filter method is applied to the use of Sobel The region near the edge is extracted by the filtering method, and the edge is extracted (step S37), and the edge is used as the edge of the object contained in the target region. Thus, by combining the Sobel filtering method and the Canny filtering method, the extraction speed of the edge becomes about 10 times as compared with the case where only the Canny filtering method is used.

Then, in step S38 of FIG. 8, the control device 1 extracts a bright object having a contour and an edge and having a contour and an edge overlapping each other and substantially matching each other as a candidate object. FIG. 27 is a view showing an example of a candidate object of the bright object extracted in step S38 of FIG. 8. Then, in step S39 of FIG. 8, the control device 1 extracts a dark object having a contour and an edge and having a contour and an edge overlapping each other and substantially matching each other as a candidate object. FIG. 28 is a view showing an example of a candidate object of a dark object extracted in step S39 of FIG. 8.

Fig. 29A is a view showing an example of an object. FIG. 29B is a diagram showing the binarized image of FIG. 29A extracted using the threshold 200. Fig. 29C is a view showing the outline of Fig. 29B. Figure 29D is a diagram showing the edge of Figure 29A extracted using threshold 50. Figure 29E is a diagram showing the edge of Figure 29A extracted using the threshold 200. Figure 29F is a diagram showing the outline and edges of a character. The object of FIG. 29A includes, for example, a portion of luminance 0, a portion of luminance 128, and a portion of luminance 255 when the luminance of the pixel varies from 0 to 255. The outline of the object of Fig. 29A is obtained as the outline of its binarized image (Fig. 29B) (Fig. 29C). The edge of the object of Fig. 29A is obtained as a part of sudden change in luminance, and different edges are extracted by using different threshold values (Fig. 29D, Fig. 29E). As shown in Figure 29C~29E Generally, the outline and edge of the object are not necessarily the same. However, it is considered that the object of the character always has a closed edge, and the contour and edge of the object are identical. Therefore, an object of a character can be extracted by extracting an object having a substantially uniform contour and an edge. Objects whose edges and contours are inconsistent are deleted as noise.

Then, in step S5 of FIG. 6, the control device 1 determines whether or not the extraction of the candidate object is successful. If YES, the process proceeds to step S6, and when NO, the process proceeds to step S10. In step S6, the control device 1 performs OCR (Optical Character Recognition) processing.

Fig. 9 is a flow chart showing the subroutine of the OCR processing in step S6 of Fig. 6. Since it is impossible to know whether the character string of the recognition object is a clear object or a dark object, and it is impossible to know whether the character string of the recognition object extends parallel to the X-axis of FIG. 5 or parallel to the Y-axis, for all of them The combination of the above performs the OCR subroutine of FIGS. 10 and 11. When the character string of the hypothetical recognition device is a bright object, the candidate object of the bright object extracted in step S38 of Fig. 8 is used. When the character string of the hypothetical recognition device is a dark object, the candidate object of the dark object extracted in step S39 of Fig. 8 is used. The image of the target area is directly used when the character string of the recognition device is extended in parallel with the X axis. When the character string of the recognition device is extended in parallel with the Y-axis, the image of the target region is rotated by 90 degrees for use.

In step S51 of Fig. 9, the control device 1 assumes that the character string of the recognition device is a bright object extending in parallel with the X-axis, and executes the OCR secondary routine. In step S52, the control device 1 determines whether the OCR is successful, and proceeds to step S7 of Fig. 6 when YES, and proceeds to step S53 when NO. In step S53, the control device 1 assumes that the character string of the recognition device is a bright object extending in parallel with the Y-axis, and executes the OCR secondary routine. In step S54, the control device 1 determines whether the OCR is successful, and proceeds to step S7 of Fig. 6 when YES, and proceeds to step S55 when NO. In step S55, the control device 1 assumes that the character string of the recognition device is a dark object extending in parallel with the X-axis, and executes the OCR secondary routine. In step S56, the control device 1 determines Whether or not the OCR is successful is successful, and when YES, the process proceeds to step S7 of FIG. 6, and when NO, the process proceeds to step S57. In step S57, the control device 1 assumes that the character string of the recognition device is a dark object extending in parallel with the Y-axis, and executes the OCR secondary routine, and thereafter proceeds to step S7 of FIG.

Fig. 10 is a flow chart showing the first part of the OCR subroutine in steps S51, S53, S55, and S57 of Fig. 9. Fig. 11 is a flow chart showing a second part of the OCR subroutine in steps S51, S53, S55, and S57 of Fig. 9.

In step S61 of FIG. 10, the control device 1 performs marking of the candidate object, and extracts a plurality of objects extending in a predetermined direction and adjacent to each other as character string candidates. Fig. 30 is a view showing an example of character string candidates extracted in step S61 of Fig. 10; Fig. 31 is a view showing the extraction of character string candidates in step S61 of Fig. 10. In FIG. 31, the 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 disposed at a certain position also includes pixels of one candidate object, it is determined that the region in the character string candidate mask 31 is one of the character string candidates. The character string candidate mask 31 is scanned over the entire target area, and labels are assigned to the individual character string candidates.

When the candidate object is marked in step S61 and the character string candidate is extracted, there is a possibility that a plurality of adjacent character strings are extracted as one character string candidate. Therefore, the character string candidates are temporarily separated from the character candidates, and the character candidates having similar feature amounts are recombined as character string candidates based on the feature amounts (width and height) of the character candidates. In step S62 of FIG. 10, the control device 1 performs marking of the object in each character string candidate, and extracts a plurality of character candidates included in the character string candidate, thereby generating a bounding box of each character candidate. Each of the bounding frames is 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 each of which surrounds the minimum circumscribed rectangle of each character candidate. In step S63, the control device 1 deforms each of the bounding frames so as to increase the width of the bounding frame as the height of the bounding frame decreases as the width and height. In step S64, the control device 1 extracts the bounding box connected by the deformation. The group of text candidates is included as a new string candidate. Fig. 32 is a view showing an example of a character candidate extracted in step S62 of Fig. 10 and a boundary block generated.

Fig. 33A is a view showing an example of character string candidates extracted in step S61 of Fig. 10; Fig. 33B is a view showing an example of the character candidates extracted in step S62 of Fig. 10 and the resulting bounding block 42. Fig. 33C is a view showing an example of the bounding frame 43 of the deformation in the step S63 of Fig. 10. Fig. 33D is a view showing an example of a new character string candidate extracted in step S64 of Fig. 10. Although the character string candidate of FIG. 33A includes two character strings "2012.1" and "abc", it is extracted as one character string candidate. For the sake of explanation, the bounding box 41 of the character string candidate is shown in FIG. 33A. Then, as shown in FIG. 33B, the object object in the character string candidate of FIG. 33A is marked, and a plurality of character candidates included in the character string candidate are extracted, and a bounding box 42 for each character candidate is generated. The bounding box 42 of each character candidate has a width w3 and a height h3. Then, as shown in Fig. 33C, the bounding frames are deformed in accordance with the width and height thereof. The width w3' and the height h3' of the bounding frame 43 after the deformation are obtained by the following formula.

[Number 8] h 3' = h 3

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 frame of each character candidate is deformed so as to increase the width w3 as the height h3 is lower. Therefore, the distance between "." and "1" is greater than the distance between "1" and "a". However, in the boundary box 43 after the deformation, "." and "1" are linked, and "1" and "a" are separated. . As shown in FIG. 33D, the character candidates included in the bounding box linked by the deformation are extracted. The group is used as a new string candidate. For the sake of explanation, the new character string candidate bounding boxes 41a, 41b are shown in Fig. 33D.

Fig. 34A is a view showing another example of the character string candidates extracted in step S61 of Fig. 10; Fig. 34B is a view showing another example of the character candidates extracted in step S62 of Fig. 10 and the resulting bounding block 42. When the group of character candidates included in the bounding box connected by the deformation in step S64 is not extracted as a new character string candidate, the object included in the character string candidate of FIG. 34A is used as noise. delete.

Then, in steps S65 to S67, in order to prevent erroneous recognition, a character string candidate that is not a character string of a date is deleted.

In step S65 of Fig. 10, the control device 1 deletes the character string candidates including more than ten character candidates. The date string is considered to contain at most 10 characters. Therefore, the character string candidate including more than ten character candidates is deleted as noise.

Then, in step S66, the control device 1 deletes the character string candidates including only the character candidates including two or more objects in the height direction. Here, for each character candidate, the object included in the character candidate is 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, although it is a character string of a date, it is considered that there is a possibility that a character candidate including two or more objects in the height direction is present due to noise or the like (in the case where there is an extra object, the object to be connected) When the object is cut, the character string candidate including only the character candidates including two or more objects in the height direction is deleted as noise. Fig. 36A is a view showing an example of character string candidates extracted in step S64 of Fig. 10; FIG. 36B is a view showing the number of objects in the height direction of each character candidate included in the character string candidate of FIG. 36A. The number of objects in the height direction of each character candidate is indicated on the character candidate 51 of Fig. 36B.

37A is a view showing another example of the character string candidates extracted in step S64 of FIG. Figure. 37B is a view showing the number of objects in the height direction of each character candidate included in the character string candidate of FIG. 37A. 37A and 37B are examples in which the character string extends in the longitudinal direction, but the erroneous processing is extended in the lateral direction. 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 of FIGS. 37A and 37B are deleted as noise.

Then, in step S67, the control device 1 selects a region of the character string candidate that is 60% or less of the area (pixel number) of the pixel of the edge from the portion where the pixel at the edge of each object is removed from the contour pixel. When the candidate object of the bright object or the dark object is extracted, only the correct character string candidate is included in one of the objects. It is considered that in the candidate object that is correctly extracted, the pixels at the edges substantially coincide with the pixels of the outline. On the other hand, the character string candidate extracted from another candidate object is deleted as noise. As the edge of the object, the edge extracted in step S37 of Fig. 8 is used. Fig. 38A is a view showing an example of an input image. Fig. 38B is a view showing a candidate object of the bright object extracted from the image of Fig. 38A. Fig. 38C is a view showing the outline of the candidate object of Fig. 38B. Figure 38D is a diagram showing the edges extracted from the image of Figure 38A. Fig. 39A is a view showing an example of an input image. Fig. 39B is a view showing a candidate object of a dark object extracted from the image of Fig. 39A. Fig. 39C is a view showing the outline of the candidate object of Fig. 39B. Figure 39D is a diagram showing the edge extracted from the image of Figure 39A. The input image of Figures 38A and 39A contains dark objects. Thus, although the outline of FIG. 39C is substantially identical to the edge of FIG. 39D, the outline of FIG. 38C is inconsistent with that of FIG. 38D.

FIG. 35 is a view showing an example of character string candidates after a part of the character string candidates are deleted in steps S65, S66, and S67 of FIG. Compared with Fig. 30, it is known that the noise has been reduced.

Then, in step S68 of Fig. 11, the control device 1 performs edge intensity and area brightness determination processing.

Fig. 12 is a flow chart showing the subroutine of the edge intensity and area brightness determination processing in step S68 of Fig. 11.

In step S90 of Fig. 12, the control device 1 selects one character string candidate. In step S91, the control device 1 extracts and enlarges the outline of the character candidate region. Here, it is expanded by adding 1 pixel to the pixel of the outline of the character candidate region. In step S92, the control device 1 applies the Canni filter method to the contour expanded in step S91, and detects the edge of the region of the character string candidate. Here, a Canny filtering method having a Gaussian function standard deviation σ=1, a high threshold Th_H=30, and a low threshold Th_L=10 is used.

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

[Number 9] edge_L = edge_M -min (15, edge_D ) × 1.2

[Number 10] edge_H = edge_M + edge_D × 2

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

[Formula 11] I_L = I_M -min (15 , I_D) × 1.2

[Formula 12] I_H = I_M + I_D × 2

In step S95, the control device 1 selects one of the selected character string candidates. In step S96, the control device 1 calculates an average value of the edge intensities of the regions of the selected character candidates. In step S97, the control device 1 calculates the average value of the brightness of the region of the selected character candidate. In step S98, when the selected character candidate has the edge intensity and the brightness outside the reference range, the control device 1 deletes the character candidate. In detail, a character candidate having an edge strength that does not reach the lower limit edge_L or an edge intensity that is greater than the upper limit edge_H is deleted as noise. Further, the character candidate system having the luminance that is less than the lower limit I_L or the luminance greater than the upper limit I_H is deleted as noise.

Then, in step S99, the control device 1 determines whether 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 there is an unprocessed character string candidate. If YES, the process proceeds to step S90, and when NO, the process proceeds to step S69 of FIG.

Fig. 40 is a view showing an example of an input image for explaining the edge intensity and the area brightness determination processing in step S68 of Fig. 11. Fig. 41 is a view showing an example of character string candidates selected in step S90 of Fig. 12; Fig. 42 is a view showing an example of character string candidates processed by the edge intensity and region luminance determination processing in step S68 of Fig. 11; As can be seen from FIGS. 40 to 42, the noise is reduced based on the edge intensity and the area brightness.

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

Fig. 13 is a flow chart showing the subroutine of the average height determination processing in step S69 of Fig. 11. In step S101, the control device 1 selects one character string candidate. In step S102, the control device 1 connects the objects included in the character candidates in the height direction. In order to connect the object in the height direction, the sealing process is performed in the height direction (that is, the expansion process of the region is performed, and then the shrinkage process is performed). In step S103, the control device 1 calculates the average value and the deviation of the heights of the character candidates, and determines the reference range of the height. In order to determine the reference range of the height, the intermediate value of the height of each object in the character string candidate may be calculated instead of calculating the average and the deviation. In this case, for example, it may be more than 5 pixels in height. A range of 1.1 times or less of the inter-value is used as a reference range. In step S104, the control device 1 deletes the character candidates that do not have the height of the predetermined range, and extracts new character string candidates separated by the deletion of the character candidates from the original character string candidates. Fig. 43A is a view showing an example of character string candidates selected in step S101 of Fig. 13; Fig. 43B is a view showing an example of a new character string candidate extracted in step S104 of Fig. 13; The character candidate 42a is deleted as noise, and new character string candidates 41c and 41d are extracted. Then, in step S105 of 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 when NO, the process proceeds to step S70 of Fig. 11.

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 process proceeds to steps S52, S54, and S56 of Fig. 9 or step S7 of Fig. 6, and when 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 the subroutine of the date pattern determination processing in steps S72 and S75 of Fig. 11.

The control device 1 holds therein a table including a plurality of date patterns indicating two or four digits of the year, one or two digits representing the month, and a predetermined punctuation mark. If "July 2012" is taken as an example, the date has, for example, the following pattern.

Each date pattern is in which the number and punctuation are specified.

In step S111 of FIG. 14, the control device 1 performs alphanumeric use for character string candidates. OCR. In step S112, the control device 1 selects one of 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 coincides with the date pattern selected in step S112. In step S113, the control device 1 determines whether or not the character string and the date pattern match. 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 "English characters that are clearly not mistaken for the number". If YES, the process proceeds to step S115, and if NO, the process proceeds to step S117. "It is obvious that the English text of the number is not mistaken." For example, "O", "o", "C", "c", "U", "u", "Z", "z", "n", " L", "l", "I", "J", "D". When the string contains "English text that is obviously not mistaken for numbers", it is recognized that the string is not a date. In step S115, the control device 1 determines whether or not all of the date patterns have been used. If YES, the process proceeds to step S73 of FIG. 11 (or step S76). If NO, the process returns to step S112, and other date patterns are selected. In step S117, the control device 1 determines whether or not the height of the character in the character string is fixed. 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 the character string contains the English character, and 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 digital OCR on the character string candidate, and recognizes the character string candidate as a number even if it contains English characters.

After step S117, tilt correction can also be performed. The string containing "1" can be recognized without any mistake by performing the tilt correction.

In step S118 of Fig. 14, the control device 1 determines whether the character string is the date, and if YES, proceeds to step S73 of Fig. 11 (or step S76), and returns to step S113 when NO.

In step S73 of Fig. 11, the control device 1 judges whether or not the determination of the date pattern is successful. If YES, the process proceeds to steps S52, S54, and S56 of Fig. 9 or step S7 of Fig. 6, and when NO, the process proceeds 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 processing as the above-described character string candidate rotated by 180 degrees. In step S76, the control device 1 determines the determination of the date pattern. If it is successful, the process proceeds to step S52, S54, S56 of Fig. 9 or step S7 of Fig. 6 at the time of YES, and proceeds to step S77 when NO. 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, the process proceeds to steps S52, S54, and S56 of FIG. 9 or step S7 of step S7 of FIG.

In step S7 of Fig. 6, the control device 1 determines whether the OCR is successful, that is, whether the extraction of the character string indicating the date of the expiration date is successful, and proceeds to step S8 when YES, and proceeds to step S10 when NO. When it is found that the month is "1", there is a possibility that it is actually "10" to "12", but it is erroneously recognized as "1" due to the angle of the container 13. In the following procedure, when the character string candidate of one input image includes only "1" as the number indicating the month, it is determined whether or not the character string candidate of the other input image includes only "1" as the number indicating the month. In step S8, the control device 1 determines whether the month is "1", and proceeds to step S9 when YES, and proceeds to step S12 when NO. In step S9, the control device 1 determines whether or not the same date has been detected twice, and if YES, the process proceeds to step S12, and when NO, the process proceeds to step S10. In step S10, the control device 1 determines whether or not the container 13 has been rotated once, and proceeds to step S13 when YES, and proceeds to step S11 when NO. In step S11, the control device 1 rotates the container 13. For example, when the container 13 is rotated by 15 degrees each time, a total of 24 input images can be obtained. Further, in order to obtain an image corresponding to each fixed angle even when the diameter of the container 13 is different, the container 13 can be rotated for two weeks and the diameter can be detected, and the image of the container 13 can be obtained at a fixed time interval. The container 13 is rotated at a different speed depending on its diameter. In step S12, the control device 1 outputs the date. In step S13, the control device 1 outputs an error.

As described above, according to the optical character recognition apparatus of the present embodiment, the various noises contained in the image of the character string can be removed in advance by optically recognizing the character string, and the representation can be expressed higher than the previous precision. A string of dates.

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

When the control device 1 is connected to the external PC 9, the date detection processing of FIGS. 6 to 14 can also be performed at least partially by the PC 9.

As described above, an optical character recognition method for recognizing a character string indicating a date can also be implemented. Moreover, such an optical character recognition method can also be implemented as a computer program that optically recognizes a character string when executed by a computer. Moreover, such a computer program can also be stored in a recording medium that can be read by a computer. For example, when such a computer program is stored in the recording medium 10 of FIG. 1, and the PC 9 reads the computer program from the recording medium 10, the optical character recognition method is implemented in accordance with the computer program.

Second embodiment.

The optical character recognition device according to the first embodiment can recognize a character string indicating a date higher than the previous precision as long as the date includes a predetermined punctuation mark and is printed in a character having a normal font. However, in order to identify the date (such as "2015/5", "2015 5") containing special punctuation marks, the date of printing with a special font (for example, a number of points separated from each other), etc. ( Hereinafter, it is called a special pattern), and it is necessary to widen various conditions (threshold values, etc.) in the date detection processing of FIGS. 6 to 14 and extract a plurality of character string candidates. Since the condition for the judgment in the date detection processing is relaxed, the amount of noise increases, and the time required for the execution date detection processing becomes long. Therefore, it is desirable to recognize the date of the special pattern while suppressing the increase in the execution time.

Fig. 44 is a flowchart showing the 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 of Fig. 44 are the same as steps S1 to S12 of Fig. 6. The date detection processing of Fig. 44 includes the date detection processing of the special pattern of step S14 instead of step S13 of Fig. 1. In the date detection processing of the special pattern in step S14, conditions (threshold values, etc.) which are more relaxed than the users in steps S2, S4, and S6 of Fig. 44 are set, and the date detection processing of Fig. 6 is executed. The types and quantities of containers on which the date of the special pattern is printed are small, and the date printed on most of the containers can be identified by performing a date detection process that is set to some extent with a limited condition.

According to the date detection processing of FIG. 44, since the date detection processing of the special pattern of step S14 is executed only when the character string of the date cannot be recognized by the steps S1 to S11 of FIG. 44, it is possible to recognize the special while suppressing the increase of the execution time. The date of the pattern.

Third embodiment.

According to the optical character recognition device of the first embodiment, since a plurality of images (input images) respectively indicating different angles of the container 13 are obtained, when the character string of the date is orthogonal to the rotation axis of the cylindrical container ( Figures 3 and 4) show the possibility that the entire date is not included in one image. When the character string of the date is over half a week of the side of the cylindrical container, the image including the entire date cannot be obtained.

Fig. 45 is a flowchart showing the 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 of Fig. 45 are the same as steps S1 to S12 of Fig. 6. The date detection processing of Fig. 45 includes the date detection processing of the concatenated image of step S15 instead of step S13 of Fig. 1. Fig. 46 is a flow chart showing the subroutine of the date detection processing of the concatenated image in step S15 of Fig. 45. As described above, the control device 1 photographs the container 13 by the camera 5 while rotating the container 13 by a fixed angle with the rollers 8a and 8b, thereby obtaining a plurality of images respectively indicating different angles of the container 13. In step S1A of Fig. 46, the control device 1 connects a plurality of images including the mutually adjacent portions of the container 13 to generate one connected image. Specifically, the control device 1 connects the images by recognizing similar objects among the two adjacent images. The connected image is a planar image of the side of the container. The control device 1 corrects the width of the container 13 as the diameter of the cylinder with respect to the curved portion of the curved surface of the container 13, and corrects it in a plane using projective transformation. Steps S2 to S9, S12 and S13 of Fig. 46 are the same as steps S2 to S9, S12 and S13 of Fig. 6.

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

Previously, in order to produce an image of the side of a cylindrical shaped object, a wire array was known. machine. However, in order to use a line camera, in addition to the cost of the line camera itself, it is also costly to provide a mechanism for rotating the object with high precision. Since the container of the medicine has various shapes and sizes, in order to perform photographing using the line camera, the cost for rotating it with sufficient precision becomes very large. On the other hand, according to the date detection processing of FIG. 45, a plurality of images captured by a normal camera are connected to generate a connected image, whereby the increase in cost can be suppressed.

According to the date detection processing of FIG. 45, since the date of the link image in step S15 is executed only when the character string of the date cannot be recognized in steps S1 to S11 of FIG. 45, it is possible to recognize the increase in the execution time while recognizing the execution time. Date not included in 1 image.

Fourth embodiment.

When there are other characters ("O", "J", "Z", etc.) just after the string of the date (on the right side), there is a possibility that the other characters are mistakenly recognized as one of the dates. Therefore, it is necessary to remove other words that are not part of the date from the string candidates.

FIG. 47 is a flowchart showing a subroutine of the date pattern determination processing of the 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 processing of Fig. 47 is executed in steps S72 and S75 of Fig. 11, and steps S121 and S122 are added between steps S113 and S114 of Fig. 14. In step S121 of FIG. 47, when the number indicating the month of the character string matching the date pattern is "10", "11", or "12", the control device 1 judges based on the criteria described below. Whether there is another character after the date, the process proceeds to step S122 when YES, and the process proceeds to step S114 when NO.

Fig. 48 is a view showing an example of a character string candidate including a character string of a date and other characters. There is another text "CJ932" after the date "2016.1". If "C" is misidentified as "0", the character string matching the date pattern is misidentified as "2016.10". When the control device 1 detects the date string "2016.10", it judges the last "0" actually. Is it a part of the date?

The distances D1 to D10 between the characters in Fig. 48 are in units of pixels, for example, as described below.

In order to determine whether the last "0" of "2016.10" is part of the date, first consider the case where the distance between the characters of the date is D1~D6. In this case, since the distance between the last "1" and "0" (the distance between "1" and "C") D6 is the same as the distance D5 between "." and "1", "2016.10" cannot be judged. 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 one of the dates, the average value of the distances D7 to D10 between the distances D6 and the characters after "2016.10" is compared. When the distance D6 is greater than the average value of the distances D7 to D10, the control device 1 judges that the last "0" of "2016.10" is not one of the dates, and the character "CJ932" is removed in step S122. 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 one of the dates. Thereby, even if there are other characters just after the string of the date, the character string indicating the date can be recognized with high precision.

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

[Industrial availability]

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 indicating a date higher than the previous precision.

S1~S13‧‧‧Steps

Claims (14)

An optical character recognition device that optically recognizes a character string, wherein the optical character recognition device includes: a first processing unit that extracts a target region including an object to be recognized from an input image, the input image An image including a container or an image attached to the label of the container; and a second processing unit that extracts a candidate object including at least one character string candidate from the object included in the target region; The third processing unit performs marking of the candidate object, extracts a plurality of objects extending in a predetermined direction and adjacent to each other as the character string candidate, and determines whether the character string candidate includes two digits indicating the year or a four-digit number, a pattern indicating a date of one or two digits of the month, and a date of a predetermined punctuation mark, and identifying the character string candidate as a date when the character string candidate has the pattern of the date; and the input map An image of a cylindrically shaped container that is rotatably held; the optical character recognition device takes one side to rotate the container a plurality of input images respectively indicating different angles of the containers, and the third processing means determining other input images when the character string candidates of one input image include only "1" as the number indicating the month Whether the character string candidate contains only "1" as the number indicating the month. The optical character recognition device according to claim 1, wherein the second processing means detects an outline and an edge of the object included in the target area, and extracts an object having a contour and an edge overlapping each other as the candidate object. The optical character recognition device of claim 2, wherein the second processing mechanism is: Applying a Sobel filtering method to the target region to detect a first edge, applying a Canni filter method to a region near the first edge to detect a second edge, and using the second edge as an object included in the target region The edge. The optical character recognition device according to claim 1, wherein the third processing means performs: marking the object of the character string candidate, extracting a plurality of character candidates, and generating a plurality of bounding boxes, wherein the plurality of bounding boxes are respectively And a rectangular shape having a width parallel to a direction in which the character string candidate extends and a height orthogonal to a direction in which the character string candidate extends, and surrounding each of the character candidates, wherein each of the bounding boxes has a lower height of the bounding frame Then, the width of the bounding frame is expanded to be deformed, and the group of character candidates included in the bounding box connected by the deformation is extracted as a new character string candidate. The optical character recognition device according to claim 1, wherein the third processing means performs the marking of the object of the character string candidate, extracts a plurality of character candidates, and deletes a character string candidate including more than ten character candidates. The optical character recognition device according to claim 1, wherein the third processing means performs the marking of the object of the character string candidate, extracts a plurality of character candidates, and deletes only the direction orthogonal to the direction of the character string candidate extension. The character string candidate of the character candidate of two or more objects is included in the direction. The optical character recognition device according to claim 1, wherein the third processing means detects an outline and an edge of the object of the character string candidate, and deletes a portion of the pixel of the edge that matches a pixel of the contour as the edge A string candidate of 60% or less of the area of the pixel. The optical character recognition device according to claim 1, wherein the third processing means recognizes the character string candidate as a non-date when the character string candidate includes an English character that is not significantly misjudged as a number. The optical character recognition device according to claim 1, wherein the third processing means is configured to include at least one other character indicating the month and two characters subsequent to the two digits of the month, and When the distance between the two digits of the month is greater than the average of the distance between the two digits of the month and other characters and the distance between the other characters, the one digit of the two digits of the month is removed and the above Other text. The optical character recognition device of claim 1, wherein the first processing means extracts an edge extending from a direction substantially orthogonal to a rotation axis of the cylindrical container from the input image, and the brightness is higher than A region of a predetermined threshold value is used as the target region. An optical character recognition device according to claim 1, wherein said optical character recognition device comprises: a camera; a photographing table that holds said container in such a manner as to be rotatable about a rotation axis of said cylindrical container; and a moving device that enables The container moves between at least one of the storages and the imaging table; and a character string indicating a date of use of the medicine in the container is printed on the container; the optical character recognition device further includes: fourth processing The mechanism determines whether the container is stored in at least one of the storages based on the date recognized by the third processing unit. An optical character recognition method for optically recognizing a character string, wherein the optical character recognition method includes: a first step of extracting, from an input image, a target area including an object to be recognized, the input image An image including a container or an image attached to the label of the container; and a second step of extracting a candidate object including at least one character string candidate from the object included in the target region; In the third step, the candidate object is marked, and a plurality of objects extending in a predetermined direction and adjacent to each other are extracted as the character string candidates, and it is determined whether the character string candidate includes two digits representing the year or a four-digit number, a pattern indicating a date of one or two digits of the month, and a date of a predetermined punctuation mark, and identifying the character string candidate as a date when the character string candidate has the pattern of the date; and the input map An image of a cylindrically shaped container that is rotatably held; the optical character recognition method further includes taking one side of the container a step of rotating a plurality of input images respectively indicating different angles of the container; and the third step is to determine other input when the character string candidate of one input image includes only "1" as the number indicating the month Whether the character string candidate of the image contains only "1" as the number indicating the month. A recording medium, which is a computer readable recording medium storing a computer program for optically recognizing a character string when executed by a computer, characterized in that the computer program comprises: a first step, which is a self-input diagram For example, extracting a target area of an object including the recognition object, the input image comprising an image of the container or an image attached to the label of the container; In the second step, the candidate object including the object of at least one character string candidate is extracted from the object included in the target region; and the third step is to mark the candidate object and extract the object a plurality of objects extending in a predetermined direction and adjacent to each other as the character string candidate, and determining whether the character string candidate includes a 2-digit or 4-digit number indicating a year, a 1-digit or 2-digit number indicating a month, and a predetermined a pattern of a date of the predetermined punctuation mark, wherein the character string candidate is recognized as a date when the character string candidate has the pattern of the date; and the input image is an image of a cylindrical container rotatably held The computer program further includes a step of obtaining a plurality of input images respectively indicating different angles of the container while rotating the container; and the third step is that the character string candidate for one input image includes only "1" When the number of the month is displayed, it is determined whether or not the character string candidate of the other input image contains only "1" as the number indicating the month. An optical character recognition device that optically recognizes a character string, wherein the optical character recognition device includes: a first processing unit that extracts a target region including an object to be recognized from an input image; and a second processing mechanism And extracting, from the object included in the target area, a candidate object including at least one character string candidate; and a third processing unit that marks the candidate object and extracts the direction in a predetermined direction a plurality of objects extending and adjacent to each other as the character string candidate, and when the character string candidate has a pattern of a predetermined date, the character string candidate is recognized as a date; and the input image is rotatably held An image of a cylindrical container; the first processing mechanism extracting from the input image includes substantially along An edge extending in a direction orthogonal to a rotation axis of the cylindrical container and a region having a luminance higher than a predetermined threshold value are referred to as the target region.