JPH1069524A - Character segmentation method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character segmentation method capable of improving a character recognition rate and character recognition accuracy by segmenting characters so as to be the character frame of a size corresponding to the size even in the case that the sizes of the characters inside images are different. SOLUTION: A character segmentation part 44 projects a binary image in a character string direction, calculates a first reference point, specifies the area of a character string with the first reference point as a center, performs projection in a direction orthogonal to the character string direction in the specified area, calculates plural second reference points and obtains a character interval. An arithmetic expression for computing the size of the character frame based on the character interval is set to the character segmentation part 44 beforehand and the character segmentation part 44 calculates the size (column size × row size) of the character frame by substituting the character interval to the arithmetic expression, segments the character string so as to be the calculated column size with the first reference point as the center and segments the respective characters included in the character string so as to be the calculated row size with the second reference point as the center.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for picking up characters such as numbers and letters attached to an article, and cutting out the characters in the image one by one when recognizing the characters in the obtained image.

[0002]

2. Description of the Related Art In the manufacturing industry, lot numbers and serial numbers are directly printed on intermediate products and products, or stickers on which these numbers are printed are attached, and an image pickup device provided with a character recognition device is manufactured or installed on a manufacturing line. It is installed in a retrieval line or the like, reads characters attached to intermediate products or products by the imaging device, recognizes a lot number or production number from the read characters, and sends the recognized lot number or production number to a management computer. The computer performs centralized control of the manufacturing process and inventory. Characters such as lot numbers and serial numbers are often printed directly on intermediate products or products, but the printing quality is low due to blurring, bleeding, dirt, and the like. In addition, the illuminance is low and the imaging environment is poor. On the other hand, the character recognition in the manufacturing industry is required to recognize characters at high speed, at a high rate, and with high accuracy. Therefore, the following character recognition method has been adopted to meet these requirements.

A recognition target character attached to an object to be imaged is imaged by an imaging device arranged directly above, and an analog signal output by the imaging device is converted into a digital signal of a predetermined gradation, which is stored in an image memory. The character image stored in the image memory is 2
The binarized image is generated by binarization, and characters are cut out from the binarized image one by one in a character frame of a preset size.
The extracted characters are subjected to pattern matching with a plurality of reference character patterns registered in advance in the dictionary, and the degree of coincidence between each reference character pattern and the extracted characters is calculated based on the following equation (1). If the first candidate character satisfies both of the following expressions (2) and (3), the extracted character is the first candidate character and the second candidate character in order from the one with the highest matching degree. If it is determined that the character is a candidate character and only one of the two expressions is satisfied, or if neither is satisfied, the extracted character is determined to be unreadable. Matching degree = (total number of matched pixels) / (total number of pixels forming reference character pattern) (1) Matching degree of first candidate character ≧ first threshold (2) Matching degree of first candidate character Difference from matching degree of two candidate characters ≧ second threshold value (3)

In such character recognition, when cutting out characters, it is important that the cut-out characters and the character frame be well-balanced.

FIGS. 7 and 8 are explanatory views for explaining a conventional character cutting method. As shown in FIG. 7A, “1, 2, 3, 4, 5” and “6, 7, 8, 9, 9,
When there is a character string of "0", the operation of accumulating the respective binarized values forming the binarized image in the character string direction is repeated to generate a projection in the character string direction, as shown in FIG. 7B. Two peaks are obtained according to the character string, and the center of each peak is obtained and used as a first reference point for cutting out, and both peaks are set so as to have a preset column size around each first reference point. Cut out a character string.

For example, "1, 2, 2, 3"
For 3,4,5 ", 2 in the direction orthogonal to the character string direction
By repeating the operation of accumulating quantified values to generate projections,
As shown in FIG. 8B, five peaks corresponding to each character are obtained. Then, the center of each peak is obtained and set as a second reference point, and each character is cut out so as to have a predetermined line size centered on each second reference point, thereby obtaining "1" as shown in FIG. The characters ",""2,""3,""4," and "5" were cut out. In addition, the same processing as described above is performed on the character string of “6, 7, 8, 9, 0” to obtain “6”,
The characters "7", "8", "9", and "0" were cut out.

[0007]

However, in the conventional method, the following problem has arisen due to the difference in the distance between the object to be imaged and the imaging device.

FIG. 9 shows a state in which an object is imaged.
FIG.₁ , S_Two , S_Three Is the object to be imaged
You. Object S₁ , S_Two , S_Three Of different heights
In FIG. 9, the object S_Two Height <imaged
Thing S_Three Height <Object S₁, Height. In addition, each
Object S₁ , S_Two , S_Three The same size
The characters to be recognized are printed. Object S₁ , S
_Two , S_Three Is transported in the direction of the arrow by the transport device 1.
The transport device 1 is located downstream of the transport direction.
An image pickup device 21 for picking up a recognition target character is provided with a support (not shown).
It is supported and fixed by a holding member. And mobile devices
1, the object S to be imaged₁ , S_Two , S_Three Characters to be recognized
When transported to a position directly below the imaging device 21, the imaging device 21
Object S ₁ , S_Two , S_Three I will capture the characters to be recognized
Swelling.

FIG. 10 is an explanatory diagram for explaining a result obtained by extracting characters from an image obtained by imaging the characters to be recognized of the objects S ₁ , S ₂ , and S ₃ by the imaging device 21 shown in FIG. the case of the captured (a) is imaged object S ₁ shown in FIG. 9, the case of the captured (b) is imaged object S ₂ shown in FIG. 9, illustrating (c) in FIG. 9 shows the case of the captured object to be imaged S _3, respectively. In the figure, “1”, “2”,... Indicate characters cut out from the image, and solid lines surrounding each character indicate character frames.

When images of objects having different heights are picked up from a fixed height, the size of characters in the image is such that the shorter the distance from the image pickup device to the object, that is, the higher the height of the object. As large. Therefore, when each character is cut out in a character frame of a fixed size, the character is overlapped with the character frame as shown in FIG. 10A, or as shown in FIG. Sometimes the size is small. Note that FIG.
In the case (c), the size of the character is appropriate for the size of the character frame. On the other hand, since the reference character pattern used as the reference for character recognition is registered at an appropriate size for the size of the character frame, in the case of FIGS. 10 (a) and (b),
The degree of matching with the reference character pattern decreases, and the character recognition accuracy and the character recognition rate decrease.

The present invention has been made in view of the above circumstances, and has as its object to project a plurality of peaks by projecting each character string in an image in a direction orthogonal to the direction of the character string. Calculate the distance between each peak, determine the size of the character frame from the calculated distance, and cut out each character pattern so that it becomes a character frame of the determined size, even if the size of characters in the image is different, that size An object of the present invention is to provide a character cutout method that can cut out a character in a character frame having a size corresponding to the character size and improve the character recognition rate and the character recognition accuracy.

[0012]

According to the character extracting method of the present invention, an image is obtained by capturing one or more columns of target characters, and a plurality of character patterns included in one or more character strings in the image are obtained. In the method of projecting in the character string direction and the direction orthogonal to the character string to obtain a peak according to the shape of the character pattern, and based on the obtained peak, cut out each character pattern so that a character frame of an appropriate size is obtained. Each string is
A plurality of peaks are obtained by projecting in the direction orthogonal to the direction of the character string, the distance between the peaks is calculated, the size of the character frame is determined from the calculated distance, and each character frame is determined to have a predetermined size. Character patterns are cut out.

FIGS. 3 to 6 are explanatory diagrams for explaining a character segmentation method according to the present invention. Now, it is assumed that a recognition target character having the same size is printed on three objects having different heights and is imaged, and three images as shown in FIGS. 3A, 3B, and 3C are obtained. I do. Each image is composed of a plurality of pixels. As shown in FIG. 3, each image has "1, 2,
3, 4, 5 "are displayed, and the relative size of each character constituting the character string is (b) <(c) <
(A).

As shown in FIGS. 4 (a), 4 (b) and 4 (c), in a predetermined area including each character string, each character string is projected in a direction orthogonal to the direction of the character string, and a plurality of peaks corresponding to each character are projected. Get. Then, the central position of each peak is obtained, and the average value of the number of pixels between adjacent central positions is calculated to obtain the character spacing. For example, in FIG. 5A, the character interval is 44 pixels, in FIG. 5B, the character interval is 40 pixels, and in FIG. 5C, the character interval is 42 pixels.

The character spacing calculated in this way is correlated with the size of characters in an image when the distance between the imaging device and the object to be imaged is different. Therefore, using a sample material, a relational expression between the character spacing in the image and the size of the character frame that can appropriately cut out the character is determined in advance, and the character spacing determined as described above is calculated using the relational expression. Substituting to calculate an appropriate character frame size. Although the above-mentioned relational expression varies depending on the arrangement of the imaging device and the like, for example, when the following expressions (1) and (2) are used, FIG.
The character frame size (column size × row size) from which each character of (b) and (c) can be properly cut out is column size × row size = 44 pixels × 33 pixels in FIG. In FIG. 5B, column size × row size = 40 pixels × 30 pixels, and in FIG. 5C, column size × row size
Row size = 42 pixels × 32 pixels. Column size = character spacing ... (1) Row size = 3/4 x character spacing ... (2)

When characters are cut out from each character string by using these character frames, each character is cut out in a character frame corresponding to the size of the character in the image as shown in FIGS. 6 (a), 6 (b) and 6 (c). be able to. Thereby, regardless of the difference in the height of the object to be imaged, characters in the image can be appropriately cut out, and a reduction in character recognition accuracy and a character recognition rate can be prevented.

[0017]

Embodiments of the present invention will be specifically described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a character recognition device used in the embodiment of the present invention. In FIG. 1, S denotes an object to be imaged such as a steel plate or a steel pipe. The object S is conveyed in the direction of the arrow by the conveyance device 1, and characters to be recognized such as a lot number or a product number are printed in a predetermined size on a predetermined position of the object S. is there. In the transport area of the object S, the imaging device 2 is supported by a support member (not shown) so as to face a portion of the object S on which the recognition target character is printed. Is recognized. The character image picked up by the image pickup device 2 is supplied to an analog / digital (A / D) converter 3, which converts the character image into a gray-scale image having a predetermined gradation by the A / D converter 3. An arithmetic unit (CPU) 41 gives the image data to the image memory 42 and stores it there.

The grayscale image stored in the image memory 42 is binarized by a binarization unit 43. Based on the obtained binarized image, the character cutout unit 44 cuts out characters one by one in a character frame obtained as described later, and supplies each cutout character to the preprocessing unit 45. The computer 4 has a dictionary 46 in which a plurality of reference character patterns used for pattern matching are registered. The preprocessing unit 45 cuts out a given character and compares it with a reference character pattern in the dictionary 46. The direction, size, line width, and the like of the character are normalized, and are provided to the pattern matching unit 47. The pattern matching unit 47 calculates the degree of coincidence between the normalized character and each reference character pattern in the dictionary 46 according to the following equation (1), and gives the calculation result to the determination unit 48. Matching degree = (total number of matched pixels) / (total number of pixels forming reference character pattern) (1)

The judgment unit 48 determines the reference character pattern of the dictionary 46
Among them, the first candidate character and the second candidate character in descending order of matching degree
As the complement character,..., The matching degree M of the first candidate character₁And the second weather
Complementary character matching degree M_TwoIs calculated, and one of the first candidate characters is calculated.
Severity M₁And a preset first threshold D₁And the first
The difference between the matching degree of the candidate character and the matching degree of the second candidate character (M ₁
-M_Two) And a preset second threshold value D_TwoAnd the respective
Compare.

The judging section 48 judges whether the following expressions (2) and (3) are satisfied as a result of the comparison. If it is determined that the expression is satisfied, the extracted character is determined to be the first candidate character. If it is determined and it is determined that the user is not satisfied, it is determined that the content is not read. Then, the output unit 49 outputs those results. Matching degree of first candidate character ≧ first threshold (2) Difference between matching degree of first candidate character and matching degree of second candidate character ≧ second threshold (3)

FIG. 2 is a flowchart showing a procedure for extracting characters by the character extracting section 44 shown in FIG.
The character cutout unit 44 projects the binarized image in the character string direction (step S1), calculates the center position of the obtained peak (step S2), and sets that position as the first reference point. First
When the reference point is calculated, the character cutout unit 44 specifies a character string area so that the calculated first reference point is centered at a predetermined column size (step S3). (Step S4), and peaks corresponding to a plurality of characters included in the character string are obtained. The character cutout unit 44 calculates the center position of each peak (step S5), sets the position as the second reference point, calculates the average of the distance between two adjacent reference points, and obtains the character spacing. (Step S6).

An arithmetic expression for calculating the size of the character frame based on the character interval is set in the character cutout unit 44 in advance, and the character cutout unit 44 substitutes the character interval obtained in step S6 into the arithmetic expression. And the size of the character frame (column size x
(Line size) is calculated (step S7). Then, the character cutout unit 44 cuts out the character string around the first reference point obtained in step S2 so as to have the column size obtained in step S7 (step S8). Each character included in the character string is cut out so that the line size obtained in step S7 is centered on each second reference point obtained in S5 (step S9). Thereby, regardless of the difference in the height of the object to be imaged, characters in the image can be properly cut out, and a reduction in character recognition accuracy and a character recognition rate can be prevented.

[0023]

According to the character extracting method of the present invention, characters in an image can be extracted one by one in an appropriate character frame regardless of the height of the object to be imaged. The present invention has excellent effects, such as preventing a decrease in accuracy and character recognition rate.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a character recognition device used for implementing the present invention.

FIG. 2 is a flowchart illustrating a character extracting procedure performed by the character extracting unit illustrated in FIG. 1;

FIG. 3 is an explanatory diagram illustrating a character segmentation method according to the present invention.

FIG. 4 is an explanatory diagram illustrating a character segmentation method according to the present invention.

FIG. 5 is an explanatory diagram illustrating a character segmentation method according to the present invention.

FIG. 6 is an explanatory diagram illustrating a character segmentation method according to the present invention.

FIG. 7 is an explanatory diagram illustrating a conventional character segmentation method.

FIG. 8 is an explanatory diagram illustrating a conventional character segmentation method.

FIG. 9 is a schematic diagram illustrating a state in which an object is imaged.

10 is an explanatory diagram illustrating a result obtained by extracting characters from an image obtained by capturing a recognition target character of an object to be captured by the imaging device illustrated in FIG. 9;

[Explanation of symbols]

 2 Imaging device 4 Computer 42 Image memory 44 Character cutout unit 47 Pattern matching unit 48 Judgment unit

 ──────────────────────────────────────────────────の Continued from the front page (71) Applicant 000001199 Kobe Steel, Ltd. 1-3-18, Wakihama-cho, Chuo-ku, Kobe-shi, Hyogo (71) Applicant 390023928 Hitachi Engineering Co., Ltd. 3-chome, Sachicho, Hitachi-shi, Ibaraki 2-1 (72) Inventor Yuji Adachi 4-5-33 Kitahama, Chuo-ku, Osaka City, Osaka Prefecture Inside Sumitomo Metal Industries, Ltd. (72) Inventor Kazuki Hashimoto 4-5-33 Kitahama, Chuo-ku, Osaka City, Osaka Prefecture Within Sumitomo Metal Industries, Ltd. (72) Kohei Otsuki, Inventor 4-53, Kitahama, Chuo-ku, Osaka City, Osaka Prefecture Inside Sumitomo Metal Industries, Ltd. Inside Nippon Kokan Co., Ltd. (72) Inventor Nobu Takahashi 1-chome, Kawasaki-dori, Mizushima, Kurashiki-shi, Okayama Pref. 2) Inventor Hiroyuki Hasegawa 1 Kanazawacho, Kakogawa City, Hyogo Prefecture Inside the Kobe Steel Works Kakogawa Works (72) Inventor Jin Goto 3-2-1 Sachimachi, Hitachi City, Hitachi City, Ibaraki Prefecture Within Hitachi Engineering Co., Ltd.

Claims (1)

[Claims] An image is obtained by capturing one or more rows of target characters, and projecting a plurality of character patterns included in one or more character strings in the image in a character string direction and a direction orthogonal thereto. In the method of obtaining a peak corresponding to the shape of the character pattern and cutting out each character pattern so as to form a character frame of an appropriate size based on the obtained peak, each character string is set in a direction orthogonal to the direction of the character string. To obtain a plurality of peaks, calculate the distance between each peak, determine the size of the character frame from the calculated distance, cut out each character pattern so that the character frame of the determined size. Character extraction method.