KR100977252B1 - Automatic localization method from identification numbers of slab image - Google Patents

Abstract

PURPOSE: A method for automatically extracting a management number of slab image is provided to improve a success rate of slab management number extraction stage in a change of illumination and an environment of complex background. CONSTITUTION: A brightness image is extracted from a slab image(S110). A binary image is generated from the brightness image(S120). A boundary of horizontal direction is extracted from the binary image through horizontal border extraction filter of a prewit(S130). A boundary of vertical direction is extracted to coordinate-set image through a vertical border extraction filter of a prewit(S150). A final character string is selected through appropriate function among extracted candidate strings(S170). Accuracy of separate character division border is improved(S180).

Description

Automatic extraction of control number of slab image {AUTOMATIC LOCALIZATION METHOD FROM IDENTIFICATION NUMBERS OF SLAB IMAGE}

The present invention relates to a method for recognizing a management number in a slab image. More particularly, the management number of a slab image automatically finds an upper and lower boundary of a character and recognizes the management number by using information of pixels inside the boundary. It relates to an extraction method.

In recent years, slabs produced in the process of steel mill regeneration and the like, are each distinguished by a unique control number.

This management number can prevent confusion of the produced slab by checking whether the management number of the slab sent from the host computer matches the management number described in the currently-input slab.

At this time, the confirmation of whether the management numbers match is made by the manager's naked eye.

Meanwhile, a slab management number recognition system that automatically recognizes a slab management number from a slab image is also used.

In this way, when the administrator directly checks with the naked eye and uses the slab management number recognition system, the work efficiency is improved.

The slab management number recognition system cannot be used for barcodes or radio frequency identification (RFID) because of the high heat and impact applied to slab production. This is used.

The slab image is recognized as the slab management number through two steps. The first step is to extract the slab management number area from the slab image.

Subsequently, in the second step, the individual management number character image is separated from the extracted slab management number area to recognize the individual management number.

The background of the slab image is mainly the work site of the steel mill, and the lighting is severely changed and the surrounding background is very complicated.

In this environment, it is not easy to extract the management number area from the slab image, and the step of extracting the management number area from the slab image has a great influence on the efficiency of the entire slab management number recognition system.

Therefore, it is very important to increase the success rate of the first step of extracting the management number region from the slab image.

Secondly, it uses the extracted image to make the processor correctly recognize the management number.

At this time, accuracy and processing speed vary depending on the algorithm.

If the processor does not recognize the control number before the next slab enters, or if the processor does not recognize the correct number due to lack of accuracy, it is as if the product did not perform its function.

As a result, how quickly and how accurately the processor recognizes control numbers should not be overlooked.

In general, projection image analysis is used as a digital image processing technique for extracting a string region.

This method is mainly used to separate individual characters from a string having a strong strength in a limited space and goes through the following process.

First, a color image is converted into a brightness image, and a brightness image is converted into a binary image.

Next, the binary image data is regarded as a two-dimensional matrix, and the data is added in the vertical or horizontal direction. The data value added in the vertical direction represents a vertical projection profile, and the data value added in the horizontal direction is a horizontal projection profile ( Horizontal Projection Profile).

A specific character string area is detected by detecting a rapidly changing coordinate value in the vertical projection profile and the horizontal projection profile.

Then, the character string is separated from the character in the detected area.

In this process, it is assumed that the number of characters is determined first, and that the intervals of the characters are all constant.

And it divides into the same area and sees only one letter in one area and recognizes the letter.

However, when using such a vertical projection profile method or a horizontal projection profile method in a place where the environment is complicated and the lighting changes a lot, such as a steel mill, there is a problem in that it is difficult to extract a string region.

In addition, there is a problem that can not be guaranteed that the image consisting of only a string coming in the situation where several noises are mixed.

The present invention has been made to solve the problems of the prior art as described above, the object of the present invention, especially in the steel environment, such as steel, the boundary image and the boundary which is not sensitive to the change of the lighting in a severe and complex surrounding environment Provides a method of efficiently extracting the management number area from the slab image by using the information of neighboring pixels adjacent to the image, and provides a method of automatically extracting the management number of the slab image for extracting and recognizing individual characters from a character string more quickly and accurately. will be.

To this end, the method for automatically extracting a management number of a slab image according to the present invention includes a method of extracting a management number of a slab image, comprising: a contrast conversion process of generating a contrast image from the slab image; A binary conversion process of generating a binary image from the contrast-converted image; A horizontal boundary extracting step of extracting a horizontal boundary from the binary-converted image by using a horizontal boundary extracting filter of the presuite; A coordinate setting process of creating a horizontal line and a vertical line by setting the horizontal boundary extracted image into four regions by using two straight lines crossing vertically; A vertical boundary extraction process of extracting a boundary in a vertical direction using a vertical boundary extraction filter of a pre-suite for each of the four areas in the coordinated image; A candidate string extraction process of extracting noise and extracting a candidate string using the vertical boundary extracted image and the horizontal line; A final string selection process of selecting a character string corresponding to the control number having a maximum value of a fit function using a sum of variances and a sum of averages of the candidate strings from the candidate strings; An accurate division process of improving accuracy of individual character division boundaries by comparing variance of division lines and variance of right and left division lines for accurate division in the final character string; And a vertical flipping determination process for determining whether the string is flipped up and down from the string having improved accuracy.

According to the method for automatically extracting the management number of the slab image according to the present invention, there is an effect that the success rate of the slab management number extraction step can be increased in the environment of a change in lighting and a complicated background.

In addition, according to the present invention, despite the input of different character strings, it is possible to recognize characters more quickly and to improve the overall slab management number recognition rate.

Finally, according to the present invention, due to the improved recognition rate of the slab management number recognition system, administrators can also efficiently check whether the management number of the slab sent from the host computer matches the management number described in the currently entered slab. .

Hereinafter, a method for extracting a slab management number according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

In describing the present invention, when it is determined that the detailed description of the related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted.

1 is a flowchart illustrating a method for automatically extracting a management number of a slab image according to an exemplary embodiment of the present invention.

Referring to FIG. 1, the method for automatically extracting a slab management number 100 according to the present invention includes a lightness degree extraction process (S110) for extracting a brightness level image from the slab image, and a binarization for generating a binary image from the extracted brightness level image. A horizontal boundary extraction process (S130) for extracting a horizontal boundary from the binarized image by using a horizontal boundary extraction filter of the presuited image, and a horizontal axis by dividing the horizontal boundary extracted image into four regions And a coordinate setting process of creating a vertical axis (S140), a vertical boundary extraction process (S150) of extracting a vertical boundary using a vertical boundary extraction filter of a pre-suite for each of the four areas in the coordinated image, and the vertical boundary. A candidate string extracting process (S160) of extracting noise and extracting a candidate string using the extracted image and the horizontal line; A final string selection process (S170) for selecting a string having the most accurate information from the strings, an accuracy enhancement process (S180) for improving the accuracy of individual character division boundaries for accurate division in the final string, and And a vertical flipping determination process (S190) for determining whether the string is flipped up and down from the string having improved accuracy.

A method of automatically extracting a management number of a slab image according to an exemplary embodiment of the present invention configured as shown in FIG. 1 will now be described with reference to FIGS. 2 to 12.

2 and 3 are exemplary views showing an image actually input for a method for automatically extracting a management number of a slab image according to an exemplary embodiment of the present invention.

Figure 2 is an exemplary view showing the image showing the position of the slab in the slab images and the entire image that is actually marked and input by the machine or person according to the present invention.

Referring to FIG. 2, reference numerals 201 to 206 of FIG. 2 denote management numbers marked by different marking machines. And (207) to (211) are control numbers marked by a person. In addition, 212 is an actual input image photographed so that the position of the slab and the approximate left and right upper and lower boundaries may be found.

FIG. 3 is an exemplary view showing a slab image in which management numbers of FIG. 2 according to the present invention are actually mixed with various noises.

Referring to FIG. 3, among the slab images mixed with various noises, 301 to 303 of FIG. 3 show a case in which some strings are burned by heat, and 304 to 306 are smeared in the process of taking a picture. A phenomenon has occurred, and 307 to 309 appear to be connected with letters by unnecessary thin lines. The character string is a character string corresponding to the management number.

In this way, the management number of the slab image is input slightly differently for each type of marking machine, but most have a similar shape except for a few cases.

However, as can be seen with reference to Figure 3, a variety of noises, such as heat, dust, scratches outside, mixed with a large number of the management number character is damaged.

FIG. 4 is a diagram illustrating an intensity extraction process (S110) and a binarization process (S110) of an input slab image shown in FIGS. 2 and 3 in the method 100 for automatically extracting a slab management number according to the present invention illustrated in FIG. 1 according to the present invention. This is an illustration showing an example of the resultant image according to).

First, referring to FIG. 4, reference numerals 401 and 402 are color images showing an actual slab image.

And 403 and 404 is an image showing the result of the contrast extraction process (S110) of the method 100 for automatically extracting the management number of the slab image according to the present invention.

That is, since the input slab image is color, it is required to go through the contrast extraction process (S110) in order to perform binary conversion.

Thereafter, the 403 and 404 are extracted to an image such as 405 and 406 shown in FIG. 4 through a binarization process S120.

Preferably, the binarization process S120 binarizes the slab contrast images 403 and 404 by applying an Otus method.

Preferably, binarization using the Otsu method is divided into two classes based on a threshold using a histogram of an image.

The histogram is obtained from a contrast level image having pixel values of 0 to 255 distributed in the image.

A frequency distribution table indicating how many pixels there are for each pixel value from 0 to 255 in the image.

In the graph, a threshold value T is set such that the sum of the variances of the pixel distributions from 0 to the specific threshold value T and the variances of the pixel distributions from the specific threshold values T to 255 is minimized.

Based on this, it is binarized to 0 if small and 255 if large.

Referring to FIG. 4, since 401, 403, and 405 are low-temperature slab images, and 402, 404, and 406 are high-temperature slab images, binary-converted images have a difference. .

That is, it can be seen that the high temperature slab image has more noise than the low temperature slab image when the binarization process S120 is performed.

5 is an exemplary view showing an example of a result image according to the horizontal boundary extraction process (S130) of the slab management number automatic extraction method 100 according to the present invention shown in FIG. 1 according to the present invention.

Referring to FIG. 5, 501, 503, and 505 are the low temperature slab images mentioned in FIG. 4, and 502, 504, and 506 are the high temperature slab images.

Referring to FIG. 5, 502 is a case in which decryption is impossible due to noise after the binarization process S120.

On the other hand, 501 is a case of a clean slab image with little noise.

In the slab images (501) to (504), the average value of each horizontal projection profile is indicated by a yellow line. On the basis of this, the area showing a high value is left, and the area showing a low value is erased.

Preferably, the image has a width x height.

For example, in an image having a width of 20 × 10, when all 20 pixel values are summed about the horizontal axis corresponding to each of the 10 vertical axes, one frequency distribution table is formed as shown in the red graph of 503 or 504.

The average of this frequency distribution table is the yellow line.

The class with the frequency distribution higher than the yellow line is then left, and the class with the frequency distribution lower than the yellow line is deleted.

This is the processing process using the horizontal projection profile, and is a widely used method for image processing.

When applied to the binarized image in the previous process called a horizontal projection profile, an area with many pixels is left, and an area that is not is deleted.

This allows you to find the upper and lower bounds of a string.

The above-described horizontal boundary extraction process (S130) will be described in more detail as follows.

First, the binarization horizontal projection profile and the boundary value horizontal projection profile are calculated by Equation 1 below.

Next, a product of profiles is calculated in order to find a boundary area, which is calculated by Equation 2 below.

Select the coordinates of the part that raises the product of the profile to the highest level, and select the boundary area with appropriate margin so that the string can be sufficiently preserved in the coordinates.

FIG. 6 is an exemplary view showing an example of an image in which the horizontal boundary extraction process S130 of FIG. 5 according to the present invention is completed.

Referring to FIG. 6, the slab images of (601) and (602) are unconditionally erased an area smaller than the reference value without any margin, but the slab images of (603) and (604) are somewhat above and below the character string. Space margin.

Then, the coordinate axis setting process S140 shown in FIG. 1 is performed.

Here, the reason why it is set to divide into at least one area | region in the coordinate axis setting process S140 is briefly described.

If the boundary value is extracted by setting the entire image area to one area, the letters oxidized by the high temperature may be lost. However, if the slab image is divided into plural (four) regions and the boundary value of each region is extracted, such loss can be prevented.

Therefore, the coordinate axis setting process (S140) is a process of creating a horizontal axis and a vertical axis by dividing the slab image into four zones.

This can be seen in FIG.

7 is an exemplary view illustrating an example of an image in which the vertical boundary extraction process S150 according to the present invention is completed.

Referring to FIG. 7, reference numeral 701 denotes a contrast slab image, which is one of images that find a horizontal boundary using the method described above and then cut it.

Preferably, 701 is the original image clipped by the horizontal boundary.

Reference numerals 702 and 703 are images obtained by extracting boundary values from the contrast image.

That is, the slab image of 702 is an image obtained by extracting a boundary value by setting the entire region as one region, and the slab image of 703 is a vertical boundary where vertical boundaries are extracted by setting four regions. It is an image that has undergone the extraction process (S150).

Therefore, the vertical boundary extraction process (S150) extracts the vertical boundary in each of the areas set as the four areas using the vertical boundary extraction filter of the prewart.

Next, a candidate string extraction process (S160) is performed to remove noise for the slab image that has undergone the vertical boundary extraction process (S150) and extract a candidate string.

As described above, the candidate character string extraction process (S160) uses a reference line when the four regions are divided into four images. Based on the horizontal line, components that do not meet this line are judged by noise and erased.

An example of the candidate string extraction process S160 is illustrated in FIG. 8.

8 is an exemplary view showing an example of the candidate character string extraction process (S160) shown in the slab management number automatic extraction method 100 of FIG. 1 according to the present invention.

Referring to FIG. 8, the color slab image of 801 is first divided into four regions by the intersection of the horizontal line A and the vertical line B, and then subjected to binary conversion results in a slab image of 802.

In this case, if the portion which does not meet with the horizontal line A of the slab image is regarded as noise and removed, a clean candidate group image such as 803 is obtained. Here, white components become candidate characters.

Next, with reference to the following FIG. 9, the method of combining each component using the character component of the said candidate group is demonstrated.

9 is an example showing the character component combining process (S170) shown in the slab management number automatic extraction method 100 of FIG. 1 according to the present invention.

10 is a diagram illustrating an image in which a left and right boundary is selected as a combination of candidate character components.

Referring to FIG. 9, first, a rectangular rectangle is formed by using a combination of components. If the ratio of the width and height of the rectangular rectangle satisfies a certain range, it becomes the real string among the strings in the rectangular rectangle.

In this case, referring to FIG. 11, a specific ratio of the width and the height of the rectangle is set based on a printed character string that is preset in the marking machine.

After several experiments with real data, the value of width / length is about 5 ~ 8 and it has a real string.

A specific range of length / length indicates a value of 5-8.

In addition, referring to FIG. 9, when the components are combined by components, the components are combined into a red rectangular rectangle as shown. This process continues, as shown in Figure 10 to find a candidate string that satisfies the specific range of the horizontal and vertical ratio.

Subsequently, a final string selection process S170 is performed in which a real string is found using a fitness function among the candidate strings.

Preferably, the fitting function assumes that the characters in the character string are equally spaced from each other because they are printed, and if the intervals are the same, this function is appropriately generated by applying this method.

11 is a view showing the print structure of the machine noting the management number of the slab image according to the present invention.

When the printed body is divided at equal intervals as shown in FIG. 11, each character can be obtained. The method is applied to the candidate character string as shown in FIG.

FIG. 12 is a diagram illustrating an image of a process of finding a real character string using images selected by a left and right boundary.

이 적합 함수 값이 가장 높게 나타나는 후보 문자열을 선택하면 그것이 진짜 문자열이 된다.First, as shown in FIG. 12, when the candidate strings are divided at equal intervals, characters exist only between the division lines l ₁ to l ₉ according to the assumption. In this case, the sum of _nine variances (l ₁ to l ₉ ) of pixel values on the dividing line is obtained. In addition, an average of 10 sums of pixel values in the regions 1 ₁ to ₁₀ between the dividing lines is obtained. Using the variance and the average, a fitness function is obtained as shown in Equation 3 below.

If you select the candidate string with the highest value for this fitted function, it becomes a real string.

delete

delete

Thereafter, an accurate division process S180 is performed.

Finally, for more accurate segmentation of the selected real character string image, if there is a variance smaller than the variance of the existing dividing line to the left and right of each dividing line, the dividing line is continuously moved. This process is repeated until the variance of the dividing line is smaller than the variance of the left and right of the dividing line. Thus, select the final dividing line for accurate division.

Finally, a vertical position discrimination process 190 is performed.

Referring to FIG. 11, the top and bottom of the real character string are determined according to whether the empty space is located at S ₇ or S ₄ .

In other words, if the empty space is located at S ₇ , the position is correct. If it is located at S ₄ , the position is reversed.

The above method has the advantage that the string extraction success rate is high even in a complex background, which is less sensitive to the change in illumination than the existing string extraction method by using the information of the boundary image and the neighboring pixels adjacent to the boundary image. Furthermore, even in a lot of noise, it is possible to accurately divide characters in a string and show a high recognition rate.

As described above, although the present invention has been described with reference to the limited embodiments and the drawings, the present invention is not limited to the above embodiments, and those skilled in the art to which the present invention pertains various modifications and variations from such descriptions. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined not only by the claims below but also by the equivalents of the claims.

1 is a flowchart illustrating a method for automatically extracting a management number of a slab image according to an exemplary embodiment of the present invention.

2 is an exemplary view showing slab images that are actually marked and input by a machine or a person according to the present invention.

FIG. 3 is an exemplary view showing a slab image in which management numbers of FIG. 2 according to the present invention are actually mixed with various noises.

 4 is a contrast conversion process (S110) and a binary conversion process of an input slab image shown in FIGS. 2 and 3 in the method 100 for automatically extracting a slab management number according to the present invention shown in FIG. 1 according to the present invention. It is an exemplary view showing an example of the resultant image according to S120).

5 is an exemplary view showing an example of a result image according to the horizontal boundary extraction process (S130) of the method 100 for automatically extracting the management number of the slab image according to the present invention shown in FIG.

FIG. 6 is an exemplary view showing an example of an image in which the horizontal boundary extraction process S130 of FIG. 5 according to the present invention is completed.

FIG. 7 is an exemplary view showing an example of an image in which the vertical boundary extraction process S150 of FIG. 6 according to the present invention is completed.

8 is an exemplary view showing an example of the candidate character string extraction process (S160) shown in the slab management number automatic extraction method 100 of FIG. 1 according to the present invention.

9 is an example showing the character component combining process (S170) shown in the slab management number automatic extraction method 100 of FIG. 1 according to the present invention.

10 is a diagram illustrating an image in which a left and right boundary is selected as a combination between candidate character components.

11 is a view showing the print structure of the machine noting the management number of the slab image according to the present invention.

FIG. 12 is a diagram illustrating an image of a process of finding a real character string using images selected by a left and right boundary.

Claims (9)

In the method of automatically extracting the management number of the slab image, Contrast extraction step of extracting the contrast image from the slab image; A binarization process of generating a binary image from the extracted contrast image; A horizontal boundary extracting process of extracting a horizontal boundary from the binarized image using a horizontal boundary extracting filter of the sweetener; A coordinate setting process of creating a horizontal line and a vertical line by setting the horizontal boundary extracted image into four regions by using two straight lines crossing vertically; A vertical boundary extraction process of extracting a boundary in a vertical direction using a vertical boundary extraction filter of a pre-suite for each of the four areas in the coordinated image; A candidate string extraction process of extracting noise and extracting a candidate string using the vertical boundary extracted image and the horizontal line; A final string selection process of selecting a character string corresponding to the control number having a maximum value of a fit function using a sum of variances and a sum of averages of the candidate strings from the candidate strings; An accurate division process for improving the accuracy of individual character division boundaries by comparing the dispersion of pixel values of the division line and the dispersion of pixel values on the left and right of the division line while moving to the left and right of each division line for accurate division in the final character string; Method for automatically extracting the management number of the slab image, characterized in that it comprises the up and down flipping determination process for determining whether the string is upside down from the correct divided string. The method of claim 1, wherein the horizontal boundary extraction process Method for automatically extracting the management number of the slab image, characterized in that for performing the horizontal projection profile for the slab image after the contrast extraction process and the binarization process. The method of claim 2, wherein the horizontal projection profile After calculating the sum in the horizontal direction with respect to the boundary values extracted through the horizontal boundary extraction filter, the average is calculated using the sum, the area of the image below the average is cut out based on the average, and the image larger than the average is obtained. Method for automatically extracting the management number of the slab image, characterized in that for selecting the upper and lower boundaries leaving only the region of. The method of claim 1, wherein the extracting of the candidate string is performed. The control number of the slab image is automatically selected based on the horizontal line in the coordinate setting process, selecting a component that meets the horizontal line as a candidate character, selecting and deleting those not meeting as noise, and determining left and right boundaries of the candidate group. Extraction method. The method of claim 4, wherein the left and right boundary determination Among all character strings that can be created by the combination of components selected as the candidate character, the ratio of the width and height is determined by using the ratio of the width and length of the string printed by the machine indicating the management number of the slab image. Method for automatically extracting the management number of the slab image, characterized in that. The method of claim 1, wherein the final string selection process is performed. After making the division lines of equal intervals among the candidate character strings, a fitting function is made using the sum of the variance of the pixel values on the dividing line and the average of the pixel values between the dividing lines, and the fitting function value. A method for automatically extracting a management number of a slab image, wherein the maximum candidate string is determined to be a real string. The method of claim 6, wherein the final string selection process is performed. Improve the accuracy of individual character segmentation by finding the variance of pixel values on the dividing line by moving each dividing line at the same interval to the left and right, and finding the dividing line with the smallest variance by comparing the variances between the variances. Method for automatically extracting the management number of the slab image, characterized in that. delete According to claim 1, wherein the up and down flipping determination process If the character is located in the area defined as the empty space in the predetermined format of the management number, the management number automatic extraction method of the slab image, characterized in that for rotating the character string 180 degrees.

Images