JP2998443B2 - Character recognition method and device therefor - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition method by image processing and an apparatus therefor. In particular, the present invention relates to a character recognition method and a device for performing proper character recognition even when the colors of characters in a measurement area are different from each other or the illumination is uneven.

[0002]

2. Description of the Related Art Characters marked on the surface of a member having a circular cross section (for example, a steel material such as a cylindrical coil) may include bleeding or blurring, and are drawn on a curved surface. In order to enable automatic recognition of such characters, it is important to detect and set a binarization level for reading characters, and therefore for determining shades, because illumination cannot be uniform. Become.

[0003] The conventional character recognition method and apparatus are based on the premise that the binarization level is properly set based on the density of the character to be read. As such a character recognition method and apparatus, see, for example,
Japanese Patent Publication No. 388 discloses an invention entitled "Numerical Recognition Method and Apparatus" which compares a character with a reference character one by one based on characteristic parameters such as a closed loop, an intersection, and an end point formed by a line element of the character, and performs character recognition based on an approximate value. Is disclosed. Also,
In the invention entitled "Character Recognition Method and Apparatus" disclosed in Japanese Patent Application Laid-Open No. 2-68683, a conventional reference value in which the pixel density (appearance frequency) of each character in the X-Y direction is patterned and When a character is determined based on an approximate value by comparing with a target character, a pattern matching method is adopted in which a noise portion is extracted and the remaining portion excluding the noise portion is compared.

On the other hand, regarding the binarization method, CAMP-IS
Paper in IJ Vol4 (1991) -320 (250) "Development of UO steel pipe plate marking character reading device"
Sets a relatively rough window for each character, cuts out the characters for each character, and adjusts the Y-axis direction in consideration of the density difference within the character. It is optimal to adopt the P tile method binarization means (method of setting the threshold value such that the ratio of the area of the background to the object becomes a known value P for each region) It discloses a character reading that determines a proper threshold value and matches an image binarized with an optimum threshold value with a basic character pattern. In addition, Japanese Patent Application No. 3-187434 relating to the applicant of the present application.
In the invention of the character number "character collation device and method therefor", when a clear binary image is not obtained in the process of character collation, a method of setting an optimal binarization level for each region is adopted. I have.

[0005]

Such a conventional character recognition device and character recognition method are based on the premise that binarization is performed correctly, and are intended for removing and separating large noises and the like generated due to other causes. It did not follow the bleeding of the characters, the change in color, or the change in the shading. Moreover, since the entire area for character recognition is binarized at a single and the same level, even if such a device attempts to recognize a character marked on a curved surface such as an outer surface of a cylindrical steel material, etc. There is a problem in that it is easily affected by uneven illumination and shading of the lens, and it is often impossible to obtain a clear binary image from the entire area.

The above-mentioned CAMP-ISIJ V
ol4 (1991) -320, the method of dividing the Y-axis direction into three and adopting the P tile method binarization means has a coarse mesh by division.
There is a problem that it is difficult to obtain a smooth binary image in the vertical direction.

Generally, when inspecting a plurality of character areas,
Even if the shape, color, glossiness, etc. of the characters existing in each character area are different, it is necessary to be able to set the binarization level for each area.

According to the present invention, character recognition is performed on the entire region at a single and the same binarization level, or the region is divided into two regions.
Character recognition that is performed to eliminate the drawbacks of the conventional method and apparatus of performing binarization and that can set an optimal binarization level for each area based on a density difference obtained from a density distribution. It is an object to provide a method and an apparatus therefor.

[0009]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention relates to a character recognition method for binarizing a character image and recognizing the character image. Scanning the character range in a first direction along the direction in which the characters are arranged, and determining a density distribution of the image along the first direction; (b) adjacent maxima on the determined density distribution Defining a plurality of regions on the density distribution by a position corresponding to the minimum value and the minimum value, and (c) multiplying a sum of the adjacent maximum value and the minimum value by a predetermined coefficient, (D) binarizing each of the regions according to the binarization level; and (e) a second direction orthogonal to the first direction. Repeating the steps (a) to (d) for each different position in To provide a character recognition method, including.

The present invention also relates to a character recognizing apparatus for recognizing a character image by binarizing the character image. (A) A character range including a character image to be recognized is defined by a character range along a direction in which characters are arranged. Means for scanning in the direction of 1 and calculating the density distribution of the image along the first direction; and (b) determining the density distribution by the position corresponding to the adjacent maximum value and minimum value on the obtained density distribution. Means for defining a plurality of areas above; (c) means for setting a value obtained by multiplying a sum of the adjacent maximum value and minimum value by a predetermined coefficient to a binarization level in each of the areas; (D) binarizing each of the areas according to the binarization level; and (e) for each of the different positions in a second direction orthogonal to the first direction, (a) to (b). Means for repeating step d).

In one embodiment, the density difference is a difference between adjacent local maximum values and local minimum values in the density distribution, and the area is a coordinate of the local maximum value and the local minimum value in the first direction. The optimum binarization level is defined by the absolute value of the difference between the values, and the optimum binarization level is の of the sum of the maximum value and the minimum value.
It is. The characters are marked on a curved surface such as a coil.

[0012]

In the present invention, a character range including an image of a character to be recognized is scanned in a first direction, a density distribution of the image along the scanning direction is obtained, and based on the obtained density distribution, An area having a required density difference and a binarization level for the area are set, and the area is binarized based on the set binarization level. The scanning of the character range in the first direction is performed by changing the position in the second direction, and the above-described processing is performed for each scanning. As a result, an optimal binarization level is set for each region.

[0013]

An embodiment of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a diagram showing a schematic configuration of a character recognition device according to the present invention and a connection relationship with another device. In FIG. 1, a numeral string 2 for coil identification is marked on a curved outer surface of a cylindrical coil 1. The number sequence 2 is illuminated by the illumination device 3. However, since the numeral string 2 is marked on the curved surface, each character region is hardly evenly illuminated.

A television camera 4 for photographing the numeral string 2 is arranged opposite to the numeral string 2, and a character recognition device 5 according to the present invention is connected to the television camera 4. Character recognition device 5
Stores an A / D converter 5-1 for A / D converting the output of the television camera 4, a first memory 5-2 for storing the output of the A / D converter 5-1 and a standard character pattern. The second memory 5-3, the third memory 5-4 for storing the character recognition result, the output of the first memory and the output of the second memory are sequentially taken out and compared, Is compared with the standard character pattern one character at a time, and the result is
And a control unit 5-5 for storing the data in the memory of the controller. The control unit 5-5 performs a unique operation for character recognition according to the present invention.

Next, the procedure of character recognition performed by the character recognition apparatus according to the present invention, focusing on the operation of the control unit 5-5, will be described with reference to FIG.
And FIG.

First, the television camera 4 captures an image of the numeral string 2 as a subject to obtain an input image 6 (A in FIG. 2). Subsequently, a provisional threshold value is obtained for the input image 6 by measuring the grayscale histogram, thereby binarizing the input image 6 and storing the binarized input image in the first memory 5-2. . By this processing, the input image of the numeric string 2 is displayed as an image having a higher density and an image having a lower density than the provisional threshold value. It will be composed of pixels of different densities.

Next, the peripheral distribution in the vertical and horizontal directions of the input image 6 binarized by the provisional threshold, that is, the value obtained by summing the pixel values in the vertical and horizontal directions for each column is obtained. , Determine a relatively rough character string range. This process obtains a histogram in which pixel values are projected in the horizontal and vertical directions,
This is a process of determining a character string range from the intersection of the density distribution. Since the input image 6 has already been binarized in the above, the character string range in the numeral string 2 is clarified by this processing.

Subsequently, an image of a numeric string within the character string range (B in FIG. 2) determined in step (1) is extracted from the input image 6. That is, the image other than the character string range is deleted from the input image 6, and only the image of the numeral string within the character string range is left.

For the extracted image of the character string range, an X-axis is set in the direction in which the numbers are arranged, and a Y-axis is set in a direction orthogonal thereto. Scanning is performed (B in FIG. 2), and the following processing is performed for each scan in the X-axis direction.

(A) Find the density distribution in the X-axis direction (FIG. 2)
C).

(B) Based on the density distribution obtained in (a), the local maximum value and the local minimum value of the density value in the X-axis direction are sequentially obtained, and the respective X coordinates are obtained. Here, the maximum value is a value at the top of the peak of the density distribution, and the minimum value is a value at the valley of the density distribution.

(C) Of the values obtained in this manner, attention is paid to an arbitrary adjacent maximum value Qt (X coordinate is Xt) and minimum value Qb (X coordinate is Xb), and | Xt-X
b | is the area determined by the appropriate binarization level.
Value. In this case, the binarization level is (Qt + Qb) /
k (k is an arbitrary positive integer) (D in FIG. 2).
For example, in the case of the density distribution as shown in FIG. 3, as for the pair of the first adjacent maximum value Qt ₁ and minimum value Qb ₁ , if their X coordinates are Xt ₁ and Xb ₁ , respectively, | Xt ₁ −
The binarization level for the region of Xb ₁ | is (Qt ₁ + Q
b ₁ ) / 2. In the case of a pair of the second adjacent maximum value Qt ₂ and minimum value Qb ₂ , the region defined by the absolute value | Xt ₂ −Xb ₂ | Binarization is performed with ₂ + Qb ₂ ) / 2 as a binarization level.

(D) In this manner, the character string range is subdivided vertically and horizontally into small regions, and the entire region is binarized while setting a suitable binarization level for each region.

In the above, a binarized image of a number sequence can be obtained, and peripheral distributions of pixels in the X-axis direction and the Y-axis direction are obtained for the binarized image. By sharpening the positions of the numbers one by one in this manner, the images of the respective numbers are cut out, and for each of the cut out images, the change in the image size due to the difference in the distance between the number string 2 and the television camera 4 Is corrected to match the size of a standard character pattern to be subjected to pattern matching.

The image of the number cut out above is 1
Each time, the matching degree is calculated by pattern matching with the standard character pattern stored in the second memory 5-3, and the font of the standard pattern showing the maximum matching degree is used as the recognition result.

FIG. 4 is a photograph showing the result of binarization by the character recognition device of the present invention. A-1 in FIG. 4 is a photograph of the original image of the numeral string 99556 marked on the outer surface of the coil, and A-2 in FIG. 4 is a photograph of the result of binarizing this numeral string by the character recognition device of the present invention. is there. B-1 in FIG. 4 is a photograph of the original image of the numeral string 91789, and B-2 in FIG.
Is a photograph of a digit string obtained by binarizing the digit string by the character recognition device of the present invention. (A-1) and (B-
In 1), the binding band of the coil is shown as a vertical black streak, which is removed by using an appropriate unnecessary signal removing technique. As described above, in the present invention, even if there is a fluctuation in brightness or blurring of characters in the left and right, up and down directions of the numeral string marked on the outer surface of the coil, the well-known signal processing technology is used to make the change. It is possible to obtain a clear binary image of a number sequence.

As described above, when recognizing a digit string marked on a curved surface such as a coil, a binarization level is individually set for each small area to obtain a binarized image. Recognition of numbers can be performed, which can reduce erroneous recognition caused by noise due to bleeding of numbers, uneven lighting, etc., so that erroneous transport of coils can be prevented and personnel required for monitoring can be eliminated. Can be.

Further, even when the colors of the marked numbers are different or the illumination is not uniform, the optimum binarization level can be set for each area, so that the character recognition accuracy is improved. Not only can the field of application of character recognition be expanded, but it is also possible to flexibly cope with various fields of application.

Although the recognition of a numeric string has been described above, the present invention is not limited to the recognition of numbers, but may be applied to the recognition of general patterns including letters and symbols such as alphabets and katakana. Can be. Thus, the term "character" as used in the specification of the present application should be understood to mean a general pattern including characters, numbers, symbols, and the like.

[0030]

As described above, the present invention has been described in detail with reference to the embodiments. The present invention is applicable to a case where a character is marked on a curved surface and a character marked with a different color is recognized. Even if the characters are difficult to recognize, such as when the characters are not uniformly illuminated, or if erroneous recognition is likely to occur, the character range including the extracted characters is shifted in the first direction. While scanning in the second direction, a density distribution along the scanning direction is obtained in each scan, and an optimal binarization level is set for each region determined based on the density difference of the obtained density distribution. Therefore, accurate character recognition can be performed irrespective of uneven illumination of the area where the character to be recognized is marked, and a clear image of the character to be recognized can be obtained.
Therefore, the present invention not only can improve the accuracy of character recognition, but also can prevent erroneous processing due to erroneous character recognition, and further reduce the number of people who monitor the occurrence of erroneous recognition. It has a special effect that it can be done.

[Brief description of the drawings]

FIG. 1 is a schematic diagram showing a configuration of a character recognition device according to the present invention and a connection relationship with another device.

FIG. 2 is a diagram for explaining a character recognition procedure in the character recognition device according to the present invention.

FIG. 3 is a diagram for explaining an optimal binarization level set in the character recognition device according to the present invention.

FIG. 4 is a photograph of an image of a digit string binarized by the character recognition device according to the present invention.

[Explanation of symbols]

1: coil, 2: numeric string, 3: lighting device,
4: TV camera, 5: character recognition device, Qt: local maximum, Qb: local minimum, 6: input image

Claims (2)

(57) [Claims] 1. A character recognition method for recognizing a character image by binarizing the character image, wherein (a) a character range including the character image to be recognized is shifted in a first direction along the direction in which the characters are arranged. Scanning, and determining a density distribution of the image along the first direction; and (b) a plurality of positions on the density distribution by positions corresponding to adjacent maximum values and minimum values on the obtained density distribution. (C) defining a value obtained by multiplying a sum of the adjacent maximum value and minimum value by a predetermined coefficient as a binarization level in each of the regions; (E) the steps (a) to (d) for each different position in a second direction orthogonal to the first direction; And a character recognition method comprising: 2. A character recognizing apparatus for recognizing a character image by binarizing the character image, wherein (a) a character range including the character image to be recognized is shifted in a first direction along the direction in which the characters are arranged. Means for scanning to obtain a density distribution of the image along the first direction; and (b) a plurality of positions on the density distribution by the positions corresponding to the adjacent maximum value and minimum value on the obtained density distribution. Means for defining an area; (c) means for setting a value obtained by multiplying a sum of the adjacent maximum value and minimum value by a predetermined coefficient as a binarization level in each of the areas; (E) the steps (a) to (d) for each different position in a second direction orthogonal to the first direction; A character recognition device comprising: