JPH10171923A - System and method for detecting character - Google Patents

Abstract

PROBLEM TO BE SOLVED: To finely extract character pattern even if it is a character printed in a curved surface part such as a steel sheet coil by performing long-legged filter processing and smoothing processing to inputted digital picture data and performing binarization processing to the resulting picture. SOLUTION: An image pickup means 1 such as a television camera or the like fetches an object on which a character to be recognized is printed, e.g. the image of a steel plate coil. This image is digitalized in a digitalizing processing part 2, and a picture having 256 gradation is formed. This picture is subjected to the prescribed picture processing, i.e. the long-legged filter processing and the smoothing processing in a picture processing part 3. The picture subjected to the above picture processing is converted into a binary picture in a binarization processing part 4, and a character recognizing part 5 performs character recognition processing based on this binary picture. Then, the respective results processed by the digitalizing processing part 2, the picture processing part 3, the binarization processing part 4 and the character recognizing part 5, is displayed in a display part 6 as occasion demands.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character detecting method and a character detecting apparatus for detecting a character used for recognizing a character printed on the surface of an inspection object such as a product. I do.

[0002]

2. Description of the Related Art In a steel mill or the like, a product, a semi-finished product, such as a character, a number, a symbol, etc. (hereinafter simply referred to as "letter") indicating information such as a model number due to the necessity of manufacturing, processing, or inventory management. ) Is often printed or stamped. Then, a system has been put to practical use in which such characters are captured by an imaging means such as a television camera and the like, and a binarization process is performed to extract a character pattern and automatically perform character recognition. However, simply binarizing an image captured by a TV camera can provide high-accuracy characters when the surface of a product or the like is not flat or has rust or dirt. It is difficult to recognize. In view of this, the present applicant has proposed a method of extracting a marked character pattern by applying a predetermined emphasis process to an obtained image and then performing a binarization process in Japanese Patent Application No. 2-99. And an apparatus were proposed.

[0003]

The above-described method and apparatus proposed by the same applicant as the present application are mainly intended to extract a character pattern engraved on an end face of a primary rolled steel material such as a billet. In this case, the side of the horizontally placed billet is illuminated toward the end face of the billet, and an image of a character engraved by a television camera similarly arranged on the side is captured.

On the other hand, a character pattern printed on the surface of a so-called "steel coil" formed by rolling a steel sheet which has been thinly rolled through a rolling process into a roll shape is extracted and character recognition is performed. Due to the nature of the product called steel coil, a TV camera that captures the image of the characters printed on the surface of the product is moved by a crane together with a lighting device above a large number of steel coils continuously arranged in the coil storage area. There are many cases where it can be installed. Since the rolled steel coil has a curvature on its surface, the light from the illumination is transmitted to the TV camera when viewed from a TV camera that is also positioned above, illuminating the surface of the steel coil with the lighting provided above. There is a part that reflects strongly toward. Such a portion appears as a very bright portion (hereinafter, referred to as a “highlight portion”) in the image captured by the television camera. FIG. 14 is a schematic diagram showing an image of the steel plate coil 40 captured from above by a television camera, and a hatched portion 41 in the drawing shows a highlighted portion.

On the other hand, a roll-shaped steel coil is easy to rotate, so that a portion on which a character is printed is not always exactly at a fixed position. There is a case where the above highlight portion is applied. For example, in FIG. 14, the upper part of the character “31” printed on the surface of the steel coil covers the highlight part 41. In such a case, the contrast between the character and its background greatly differs depending on the position of the character, and the image of the entire character cannot be clearly captured.
Even if such an image is subjected to the binarization processing, a character pattern cannot be satisfactorily extracted, and thus it is difficult to perform character recognition with high accuracy.

Furthermore, since the color of the steel coil itself is a gray system, when printing black characters on the surface, even if the entire character to be recognized is included in the background portion having a uniform contrast. Even in this case, sufficient contrast between the character and the background cannot be expected in the first place. Therefore, even if the characters printed on the steel coil are taken in by a television camera and subjected to emphasis processing, the printed character patterns cannot always be extracted well, and it is difficult to perform character recognition with high accuracy. .

The present invention has been made on the basis of the above circumstances, and provides a character detection method and a character detection device capable of satisfactorily extracting a character pattern even for a character printed on a curved surface portion such as a steel sheet coil. The purpose is to do so.

[0008]

According to a first aspect of the present invention, there is provided a character detecting method for performing a foot length filtering process on input digital image data and then performing a smoothing process. A character pattern is extracted by performing a process and performing a binarization process on an image obtained as a result.

According to a fourth aspect of the present invention, there is provided a character detecting apparatus for imaging an image including a character to be detected, and digitizing an image signal output from the imaging means to obtain a digital image having a predetermined gradation. Digitizing means, first, a foot-length filter processing is performed on the digital image, and then, an image processing means for performing a smoothing processing is performed, and a binarizing processing is performed on the image signal output from the image processing means. The image processing apparatus further includes a binarization processing unit that converts the image into a binary image, and a character recognition unit that recognizes the character based on the binary image or an image obtained by performing image processing on the binary image.

[0010] As described above, the input digital image data is subjected to image processing in the order of first performing a foot length filter process and then performing a smoothing process, thereby obtaining a character portion of the input image data. Even when the contrast with the background is not uniform, the character pattern can be extracted well when binarized thereafter, and the character can be recognized with high accuracy.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an apparatus for executing the character detection method of the present embodiment, and FIG. 2 is a schematic layout diagram showing a relative position of a steel coil of the present embodiment on which a character to be recognized is printed, illumination and a camera. And FIG.
FIG. 2 is an enlarged cross-sectional view showing an upper part of a steel sheet coil in an enlarged manner.

In the block diagram of FIG. 1, an image pickup means 1 such as a television camera captures an image of an article (here, a steel sheet coil) on which characters to be recognized are printed. This image is digitized by the digitizing processor 2 and, for example, 25
The image has six gradations. This image is subjected to predetermined image processing in the image processing unit 3. In the present embodiment, the image processing unit 3 performs a foot length filtering process and a smoothing process described later. The image on which the image processing has been performed is converted into a binary image by the binarization processing unit 4, and the character recognition unit 5 performs a character recognition process based on the binary image. The results processed by the digitizing processing unit 2, the image processing unit 3, the binarizing processing unit 4, and the character recognizing unit 5 are displayed on the display unit 6 as necessary.
Will be displayed.

In FIG. 2, 10 ₁ , 10 ₂ , 10
Reference numeral ₃ denotes a steel sheet coil obtained by winding a thinly rolled steel sheet into a roll, and has a diameter of about 1 to 2 m. Black letters indicating the model number, product number, etc. of the product are printed on the surface of the coiled steel coils 10 ₁ , 10 ₂ , 10 ₃ , and are arranged side by side in the coil place as shown in the figure. A television camera 11 serving as an imaging unit for capturing an image of a printed character and an illumination device 12 for the television camera 11 are provided above the many coiled steel coils.
3 so that it can be moved.

In order for the characters printed on the surfaces of the steel coils 10 ₁ , 10 ₂ and 10 ₃ to be captured by the television camera 11 above, the steel coils 10 ₁ , 10 ₂ and 10 ₃ are arranged as shown in FIG. When printing, it is desirable that the portion on which the character is printed comes to the top. However, since the rolled steel coil is easy to rotate, it may be shifted from the uppermost position. FIG. 3 shows how the image of a character changes when the image of the character printed by the television camera 11 is captured, depending on where the area where the character is printed is located.
This will be described with reference to FIG. If the area where the characters are printed on the surface of the steel sheet coil 10 is in the range shown in FIG. 3A, that is, the range almost directly below the TV camera 11, the TV camera 11 can catch the characters from directly above. At the same time, the contrast with the background is uniform throughout the printed characters. On the other hand, a region indicated by b is a highlight portion where strong light from the illumination 12 is reflected on the television camera 11.
As described in connection with FIG. 14, when a part of the area where the character is printed covers the range b, the background of that part becomes very bright, and the contrast between the character and the background varies depending on the position of the character area. It becomes uneven.

FIG. 4 shows that the surface of the steel coil has a vertical length of about 20 mm.
An example of an image captured by the television camera 11 of a character printed in an area of about 40 cm in width and about 40 cm in width is shown. In FIG. 4, a part (here, about one-third in the upper part) of the area where the characters are printed on the steel sheet coil covers the range shown in FIG. 3B, and about one-third in the upper part of the printed area. The background is very bright and the background is darker underneath. For the image in FIG.
When the binarization process is performed by setting an appropriate threshold, the binary image shown in FIG. 5 is obtained. However, in such a binary image, the outline of a character is unclear, and a character pattern cannot be extracted from such an image. This is because an appropriate threshold value for extracting a clear outline in the entire character cannot be set because the contrast between the character and the background greatly differs depending on the location.

FIG. 6 shows an image obtained by performing a general smoothing process on the image of FIG. 4 before binarization and then performing a binarization process. In the image of FIG. 6, the upper part of the character is largely missing, and the quality is insufficient for extracting the character pattern. FIGS. 7 and 8 show FIG. 4 before binarization.
Is an image obtained by performing a “foot length filter process” described later on the image and then performing a binarization process. The difference between FIG. 7 and FIG. 8 is caused by changing parameters when performing the foot length filter processing. 7 and 8
Although the clarity of the character pattern of the image shown in FIG. 7 is improved as compared with the image shown in FIG. 6, part of the character is missing in FIG. 7 and noise is generated in the background in FIG. I have. Since these become obstacles in performing character recognition, it is still not sufficient to perform character recognition with high accuracy simply by performing foot length filter processing before binarization. Note that FIG.
8 and FIG. 8, unlike FIG. 5 and FIG. 6, the character portion is white, but this is due to the execution of the foot length filter processing.

Therefore, the inventor of the present application tried to perform image processing combining foot length filter processing and smoothing processing before performing binarization processing. Here, the smoothing processing and the foot length filtering processing will be briefly described. FIG. 9 is a diagram for explaining the smoothing process performed in the present embodiment. Each of the nine small squares shown in FIG. 9 indicates a pixel. Here, the smoothing process performed on the center pixel 20 is performed by the following procedure. First, the gradation of the digitized image is changed from 0 (black) to 25 for each pixel.
It is assumed that there are 256 gradations up to 5 (white), and the gradation of each of a total of nine pixels, that is, the pixel 20 and its neighboring pixels, is averaged. Then, this average value is assigned as the gradation of the pixel 20. This process is performed on all pixels on the screen to obtain an image displaying each gradation assigned to each pixel. Such a process is a smoothing process. In an image obtained by performing such processing, the gradation gradually changes in a portion where the gradation rapidly changes in the original image.

FIG. 10 is a diagram for explaining the foot length filtering process performed in this embodiment. Note that, here, as in the case of FIG. 9, 256 gradations from 0 (black) to 255 (white) are considered. In FIG. 10, small squares finely divided in a mesh shape indicate each pixel on the screen. Among them, the vertically elongated hatched area at the center indicates that the gradation is lower than the surrounding area. Here, for the sake of simplicity, it is assumed that the gradation in the hatched area is “10” and the gradation in the surrounding area is “50”.

The foot length filtering process performed on one pixel 30 included in the hatched area is as follows. First,
Consider four pixels 31, 32, 33, and 34 that are n pixels up, down, left, and right from the target pixel 30. Here, the value of n is selected so as to be larger than the number of pixels corresponding to the thickness of the line of the character to be recognized. And calculating a value of L ₃₀ which is given by the following equation.

L₃₀= M₃₀・ K₃₀+ M₃₁・ K₃₁+ M₃₂・ K
₃₂+ M₃₃・ K₃₃+ M₃₄・ K₃₄ Where K₃₀, K₃₁, K₃₂, K₃₃, K₃₄Is the gradation of each pixel
And here, as described above, K₃₀= K₃₁= K₃₂= 1
0, K₃₃= K₃₄= 50. Also, M₃₀, M ₃₁,
M₃₂, M₃₃, M₃₄Is a predetermined coefficient for each pixel
Where M₃₀= -4, M₃₁= M₃₂= M₃₃= M
₃₄= 1. Therefore, in this case, L₃₀= 80
And this L₃₀Is assigned to the pixel 30. FIG.
Same as above for pixel 30 'located next to the 0 oblique line
To L₃₀′, M₃₀'= -4, M₃₁'=
M₃₂'= M₃₃'= M₃₄'= 1 and K₃₀'= K₃₁'=
K₃₂'= K ₃₃'= K₃₄Since '= 50, L₃₀'= 0
Becomes Hereinafter, the same calculation is performed for all pixels.
L assigned to each pixel₃₀The value of
An image with pixel gradation is displayed. Such processing is foot length
Filter processing.

FIGS. 11 and 12 are images obtained by performing the above-described combination of the smoothing process and the foot length filtering process on the original image of FIG. 4 and then performing the binarization process. FIG. 11 shows the case where the smoothing process is performed first, and then the foot length filter process is performed. FIG. 12 shows the case where the foot length filter process is performed first and then the smoothing process is performed. In the image of FIG. 11, the upper part of the character is largely missing,
Insufficient quality for clear extraction of character patterns. On the other hand, in the image of FIG. 12, the outline of the character appears clearly, and the character pattern can be extracted sufficiently well.

As described above, for an image in which the contrast between the character and the background greatly differs depending on the place as shown in FIG. 4, first, foot length filter processing is performed, then smoothing processing is performed, and then binarization processing is performed. It has been found that a binary image with sufficient quality for extracting a character pattern can be obtained by performing the processing. FIG. 13 is a flowchart showing the procedure of the first parameter setting part of the processing procedure of the character detection method of the present embodiment. When processing starts, first st
At ep10, an image of a character to be recognized by a television camera or the like is input. Then, in step 11, the parameters required for performing the foot length filter processing are set to certain values in consideration of the type of the inspection target and various situations. Next, in step 12, foot length filter processing is performed on the input image, and in step 13,
The operator determines the result. If the operator determines that the operation is insufficient, the process returns to step 11 again, where parameters are set again as necessary, and the foot length filter processing is performed again. This procedure is repeated several times, and when an appropriate parameter is found, the next step is to use the parameter. Next, in step 14, parameters required for performing the smoothing process are set to certain values. And step15
Performs a smoothing process, and at step 16, the operator determines the result. If the operator determines that the operation is insufficient, the process returns to step 14 again, where parameters are set as necessary, and the smoothing process is performed again.

Once the parameters for the foot length filter processing and smoothing processing are determined, image input, foot length filter processing and smoothing processing, and binarization are performed on each of the continuously conveyed caps. A series of processing such as processing and character recognition processing is automatically performed. However, each time the foot length filter processing and the smoothing processing are performed, the necessary parameters may be set each time.

It should be noted that the present invention is not limited to the above embodiment, and various modifications can be made within the scope of the invention. For example, in the above embodiment, the present invention is applied to the case where a character printed on a steel plate coil is recognized in an ironworks, but the present invention can be applied to various fields other than the above embodiment.

[0025]

As described above, according to the present invention,
As described above, the input digital image data is subjected to image processing in the order of first performing a foot-length filter process and then performing a smoothing process, so that, for example, light from illumination is strongly directed toward a television camera. When the contrast between the character part of the input image data and the background is not uniform throughout the entire character, such as when a part of the character is applied to the reflected part, Even if there is not a sufficient contrast difference, a character pattern can be satisfactorily extracted when binarization is performed thereafter. In particular, it is possible to satisfactorily extract a character pattern for a portion having a curvature in a portion where a character is printed, such as a steel coil, which is difficult with the conventional method.
Therefore, it is possible to provide a character detection method that enables character recognition with high accuracy for such characters.

[Brief description of the drawings]

FIG. 1 is a block diagram of an apparatus that executes a character detection method according to an embodiment.

FIG. 2 is a schematic layout diagram showing relative positions of a steel coil of the present embodiment on which characters to be recognized are printed, illumination, and a camera.

FIG. 3 is an enlarged sectional view showing an upper part of a steel plate coil in an enlarged manner.

FIG. 4 is a diagram showing an example of an image (original image) of a character printed on the surface of a steel coil captured by a television camera.

FIG. 5 is a diagram showing an image obtained by binarizing an original image.

FIG. 6 is a diagram illustrating an image obtained by performing a general smoothing process on an original image and then performing a binarization process.

FIG. 7 is a diagram showing an image obtained by performing a foot length filter process on an original image and then performing a binarization process;

FIG. 8 is a diagram showing an image obtained by performing a foot length filter process on an original image and then performing a binarization process;

FIG. 9 is a diagram illustrating a smoothing process performed in the embodiment.

FIG. 10 is a diagram for explaining foot length filter processing performed in the present embodiment.

FIG. 11 is a diagram showing an image obtained by performing a smoothing process first, then performing a foot length filter process, and then performing a binarization process.

FIG. 12 is a diagram showing an image obtained by performing a foot length filter process first, then performing a smoothing process, and then performing a binarization process;

FIG. 13 is a flowchart showing a processing procedure of a part for setting a parameter in the processing of the present embodiment.

FIG. 14 is a schematic diagram showing an image obtained by capturing a steel plate coil from above with a television camera.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Image pickup means 2 Digitization processing part 3 Image processing part 4 Binary processing part 5 Character recognition part 6 Display part 10 ₁ , 10 ₂ , 10 ₃ Steel plate coil 11 Television camera 12 Lighting device 13 Crane 20, 30, 31, 31, 32 , 33, 34 pixels

Claims (4)

[Claims] 1. Digital image data including a character to be detected is first subjected to a foot length filtering process, then to a smoothing process, and to a binarization process performed on the resulting image. A character detection method characterized by extracting a character pattern. 2. The foot length filtering process includes a step of multiplying a value obtained by multiplying a gradation of a target pixel by a first constant, and n pixels (n is an integer greater than 1) vertically and horizontally from the target pixel. A value obtained by adding four values obtained by multiplying the gradation of each of four pixels separated by a second constant having a sign different from the first constant to the gradation of the target pixel is referred to as the gradation of the target pixel. 2. The character detection method according to claim 1, wherein the process of performing the process is performed on all pixels. 3. The character to be detected is a character printed on a steel plate coil.
Or the character detection method according to 2. 4. An image pickup means for picking up an image including a character to be detected, a digitizing means for digitizing an image signal output from the image pickup means to obtain a digital image of a predetermined gradation, On the other hand, an image processing unit that first performs a foot length filter process, and then performs a smoothing process, and a binarization processing unit that performs a binarization process on an image signal output from the image processing unit to obtain a binary image. And a character recognition unit that recognizes the character based on the binary image or an image obtained by performing image processing on the binary image.