JP2010198475A - Character recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a character recognition device capable of correctly reading and recognizing a character displayed on steel. <P>SOLUTION: The character recognition device includes: an imaging means 2 for imaging a character display part B on steel A; an image processing means 3 for performing required digital image processing to a color image signal input from the imaging means 2 to recognize a character; and a displaying means 19 for displaying a recognition result of the character by the image processing means 3. The image processing means 3 decomposes an imaged color image into respective channel images of R, G and B, contrasts the G channel image with a reference character image stored in the image processing means 3 in advance to evaluate coincidence degrees of both the images. This can eliminate the influence of rust generated on the steel A and correctly recognize the character displayed on the steel A. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a character recognition device for recognizing characters displayed on the surface of an object, and more particularly to a character recognition device that optically reads and recognizes characters displayed on the surface of a steel material.

  Conventionally, after a character display unit is imaged with a camera and an input image is created by digitizing a video signal input from the camera, background difference image creation processing, binarization processing, noise removal processing, and character segmentation processing are performed. A character recognition method and a character recognition device are known that perform the processing in order and recognize the described characters (see, for example, Patent Document 1). In addition, regarding the handling of text data after character recognition is performed, the recognized text data is displayed on the display unit if it can be recognized, and a warning is displayed on the display unit if it cannot be recognized. (For example, refer to Patent Document 2).

  According to the invention described in Patent Document 1, in the background difference image creation process, an optimum binarization threshold value for setting the character white and the background black based on the density histogram of the background difference image is calculated, and the background difference image Is converted into a binarized image, so that character information can be accurately recognized even if the background of the input image has uneven density or dirt to some extent.

JP 2004-110369 A JP 2006-109141 A

  By the way, in the steel manufacturer and its distribution industry, individual steel materials are managed by reading the characters displayed on the end surfaces of the steel materials by engraving or painting, and these characters can be accurately recognized. There is a need for a character recognition device.

  However, unlike information transmission media such as paper, the surface of the steel material changes significantly due to the occurrence of rust. Therefore, the conventional character recognition device is based on the premise that only relatively slight density unevenness and dirt are generated. Even if is applied as it is, it is difficult to accurately recognize characters. In addition, the direction of characters displayed on the steel material changes depending on the mounting state of the steel material with respect to the holding member. Therefore, even if a conventional character recognition device that does not consider the direction of the character is applied as it is, the character is accurately displayed. It is difficult to recognize. Furthermore, since the number of characters and the display position of the characters attached to the steel material are determined depending on the type of the steel material, even if a conventional character recognition device that does not consider these is applied as it is, the character information can be recognized accurately. Have difficulty.

  The present invention has been made to solve the problems of the prior art, and an object thereof is to provide a character recognition device capable of accurately reading and recognizing characters displayed on a steel material. .

  In order to solve the above-mentioned problems, the present invention firstly, an image pickup means for picking up an image of a character display portion of a steel material and outputting a color image signal, and a digital image processing required for the color image signal inputted from the image pickup means. Image processing means for recognizing the character displayed on the character display section, and display means for displaying the recognition result of the character by the image processing means, the image processing means comprising the color image signal After creating a color image of the character display unit from the color image, the color image is decomposed into channel images of a red channel image, a green channel image, and a blue channel image, and the green channel image and the image processing means are stored in advance. The degree of coincidence between these images is evaluated by comparing with the reference character image. If the degree of coincidence between both images is equal to or greater than a predetermined value, the character It is determined that the character has been recognized, and the character is displayed on the display unit. When the degree of coincidence between the two images is less than the predetermined value, the display unit displays an unacceptable display. .

  Since red rust is generated in steel materials, characters displayed on the steel materials cannot be accurately extracted depending on how red rust is generated even if a color image file obtained from the imaging means is used. Comparison with is impossible or extremely difficult. In particular, when a character is represented on a steel material by engraving, red rust is generated in the character as well as in the background portion, so that it is impossible to extract the character depending on how the red rust is generated. On the other hand, if the green channel image decomposed from the color image file is used, the influence of red rust can be removed in such a case, and the characters displayed on the steel can be extracted. The character can be recognized by comparison with the reference character image.

  According to the second aspect of the present invention, in the first character recognition device, the image processing unit repeatedly compares the rotation of the green channel image and the reference character image, and evaluates the degree of coincidence of the two images. Made the configuration.

  The reference character image is stored in an upright state, that is, in a state in which the reference character image can be seen upward by the standing human eyes when displayed on the display means. On the other hand, the steel material may not be displayed in an upright state on the display means depending on the cross-sectional shape and the mounting state of the steel material on the holding member at the time of imaging. For example, when the cross-sectional shape of the steel material is a quadrangle, the direction of the characters is upward, rightward, leftward, or downward depending on how the steel material is placed on the holding member. Therefore, when the green channel image decomposed from the color image file is used as it is, the characters displayed on the steel material can be recognized only with a probability of a maximum of 25%. In contrast, when the rotation of the green channel image and the comparison of the reference character image are repeated, the green channel image is always in an upright state as the rotation is repeated, so that the comparison with the reference character image is possible. Characters displayed on the steel material can be recognized with a probability of up to 100%. The same applies to the case where the cross-sectional shape of the steel material is a circle, a triangle, a polygon other than a quadrangle, and other shapes.

  Thirdly, according to the present invention, in the first character recognition device, the image processing means includes the green channel image and the reference for each of the display positions of each of the plurality of characters displayed on the steel material. In contrast to the character image, the degree of coincidence between these images is evaluated.

  A predetermined number of characters are displayed in a predetermined arrangement on the steel material. Therefore, the characters displayed on the steel cannot be accurately recognized unless the characters are recognized for each character display position. For example, when the characters “1, 2, 3, 4, 5” are arranged in this order on the steel material, “1” in the first digit is unrecognizable, “2, 3, 4, 5”. When a character string is read, it is impossible to determine which number of characters was unrecognizable, so the entire read character string becomes a list of meaningless characters. On the other hand, for each of a plurality of characters displayed on the steel material, if the degree of coincidence of the images is evaluated for each character display position, the first digit “1” cannot be recognized. Since the character string “?, 2, 3, 4, 5” (display indicating that “?” Cannot be recognized) can be displayed on the display means, the second to fifth digits are displayed. This string can be trusted, and the management of steel materials can be facilitated.

  Thirdly, according to the present invention, in the first character recognition device, the image processing means performs, as the digital image processing, a hole filling process, an erosion process with a circular element, and a relative position for the green channel image. It was configured to perform sorting processing.

  Since relatively large irregularities are formed on the surface of the steel material, the characters and the background are likely to be defective as compared with the case of reading the characters displayed on paper or the like. Therefore, if the hole filling process and the erosion process with a circular element are performed on the green channel image, it is possible to repair the defects generated in the characters and the background, so that the accuracy of extracting the characters displayed on the steel material is improved. Can do. In addition, when characters are displayed on the surface of the steel material by engraving, the character strings are not aligned in a straight line and are liable to shift vertically and horizontally, but if the sorting process based on the relative position is performed, the characters are displayed on the surface of the steel material. Even when the columns are not displayed in a straight line, the character extraction accuracy can be increased.

  The character recognition device of the present invention extracts characters displayed on a steel material using a green channel image decomposed from a color image file, and compares the extracted characters with a reference character image to display on the steel material. Therefore, even when red rust is generated in the steel material, it is difficult to be adversely affected and the character can be accurately recognized.

It is a block diagram of the character recognition apparatus which concerns on embodiment. It is a functional block diagram of the character recognition apparatus which concerns on embodiment. It is a flowchart which shows the pre-process of the character recognition algorithm performed by the character recognition apparatus which concerns on embodiment. It is a flowchart of the process (1) performed by the character recognition apparatus which concerns on embodiment. It is a flowchart of the process (2) performed by the character recognition apparatus which concerns on embodiment. It is a figure which shows notionally the process (2) performed by the character recognition apparatus which concerns on embodiment. It is a flowchart of the process (3) performed by the character recognition apparatus which concerns on embodiment. It is a flowchart of the process (4) performed by the character recognition apparatus which concerns on embodiment.

  Hereinafter, an embodiment of a character recognition device according to the present invention will be described with reference to the drawings. As shown in FIG. 1, the character recognition device of this example includes an illuminating unit 1 that irradiates illumination light to a character display unit B of a steel material A that is a reading target, and an imaging unit 2 that images the character display unit B of the steel material. And image processing means 3 that performs necessary digital image processing on the color image signal input from the imaging means 2 and recognizes the characters displayed on the character display portion B of the steel material A.

  As the illuminating means 1, for example, an arbitrary one belonging to publicly known such as a light bulb, a fluorescent lamp, a light emitting diode, and electroluminescence can be used. In order to uniformly illuminate the character display portion B of the steel material A, it is shown in FIG. Thus, it is particularly desirable to devise, for example, a combination with a frame shape.

  As the image pickup means 2, a digital camera such as a CCD camera or a CMOS camera that outputs a color image signal is used.

  As shown in FIG. 2, the image processing unit 3 includes a camera interface 11 that performs data communication with the imaging unit 2, and a color image signal input from the imaging unit 2 via the camera interface 11. An image file unit 12 that is stored as a color image file, a reference character image storage unit 13 that stores a reference character image that serves as an evaluation criterion for characters read from the character display unit B, and an image stored in the image file unit 12 A character recognition unit 14 for recognizing the character read from the character display unit B by comparing the file and the reference character image stored in the reference character image storage unit 13, and a character recognition process executed by the character recognition unit 14 An initial setting file 15 storing various parameters necessary for the image processing and an image processing library storing various application software for image processing. 16, and a recognition result file 17 for storing the character recognition result obtained by the character recognition unit 14. In the image processing means 3 of this example, numbers from 0 to 9 are stored in the reference character image storage unit 13, and the initial setting file 15 includes the number of digits of the character string (number sequence) to be evaluated, The character recognition unit 14 evaluates that the character has been recognized, the evaluation rate that is the matching rate between the character extracted from the image file and the reference character image, and the character height, width, area, etc. A feature amount that is a value for decomposing into regions is stored.

  Further, the image processing means 3 reads out parameters corresponding to the character display part B of the steel material A, which is the reading object, from the initial setting file 15 to the character recognition unit 14, and displays the characters of the steel material A, which is the reading object. An input unit 18 for reading application software corresponding to the part B from the image processing library 16 to the character recognition unit 14 is provided. In the example of FIG. 2, the keyboard 21 and the mouse 22 are cited as the input means 18, but the gist of the present invention is not limited to this, and any other input device belonging to public knowledge may be used. Can do.

  Further, the image processing means 3 is provided with a display means 19 for displaying a character recognized by the character recognition unit 14 or an error mark when a character is not recognized by the character recognition unit 14. As the display means 19, any known display device such as a liquid crystal display, a plasma display, an organic EL display, and an LED display can be used.

  In the example of FIG. 2, the input unit 18 and the display unit 19 are built in the image processing unit 3, but the gist of the present invention is not limited to this, and any of the input unit 18 and the display unit 19 is used. One or both may be externally attached to the image processing means 3.

  Hereinafter, the algorithm of the character recognition process performed in the image processing means 3 will be described with reference to FIGS.

  FIG. 3 shows a preprocessing algorithm for initializing a log file stored in the image file unit 12 (step S1), creating an image file name stored in the image file unit 12 (step S2), and from the image file unit 12 Read the first image file, determine the aspect ratio of the image displayed on the display means 19 (step S3), restart the system (step S4), and change the required lookup table (step S5). It goes in order and shifts to the process (1) shown in FIG.

  The process (1) shown in FIG. 4 is the first half of the character recognition algorithm, and a loop for each image file is started after the process of FIG. 3 is completed (step S6). In character recognition of each image file, after initializing the arrangement of the reference character image and the arrangement of the recognized character area (step S7), one image file is read from the image file section 12 (step S8), One image file is decomposed into channel images of a red channel image (R), a green channel image (G), and a blue channel image (B), and characters are cut out using the green channel image (step S9). ). In this way, by using the green channel image, even when red rust is generated in the steel material A, the influence can be eliminated and the character can be easily cut out.

  Next, the green channel image is repeatedly rotated according to the rotation angle of the image file selected from the initial setting file 15 by operating the input means 18 (step S10). For example, when the steel material A is a square, the green channel image is rotated four times from 0 ° to 270 ° in 90 ° increments. Thereby, it is possible to determine the character string distribution and the character direction. Then, when it is determined that the character string distribution and the character direction match the character string distribution and the character direction stored in the reference image storage unit 13, the process proceeds to step S11, and the dynamic threshold value for each image unit is obtained. The determination process is executed. In this dynamic threshold determination process, the binarization of the green channel image is repeated using the binarization threshold start threshold to the end threshold selected from the initial setting file 15 by operating the input unit 18. Is done. Thereafter, the process proceeds to the process (2) shown in FIG.

  In the process (2) shown in FIG. 5, the green channel image is rotated according to the character string distribution and the character orientation determined in step S10 (step S12), and then the character is subjected to slight inclination correction (step S12). S13), the green channel image is binarized according to the dynamic threshold of the image unit determined in step S11 (step S14). Since there are relatively large irregularities on the surface of the steel material A, which is the management object, missing characters and noise are likely to appear in the image generated in step S14. Therefore, hole filling (step S15) and erosion processing with a circular element (step S16) are performed by the feature amount of the region (step S15) and circular elements are eliminated, and noise is removed. I do.

The “region feature amount” is a feature amount for interpolating a hole, and as an example, it is a minimum value and a maximum value such as area, ellipse minor axis and major axis, inscribed circle, circumscribed circle, etc. Are stored in the initial setting file 15 and are selected by operating the input means 18. By performing the filling process in step S15, as shown in FIG. 6A, even when a hole 22 appears in the character 21, it is between the minimum value and the maximum value of the feature amount of the preselected region. The size of the hole 22 is filled, and the extracted character 21 becomes more probable. In FIG. 6A, the minimum area value 1.0 mm ² and the maximum area value 10.0 mm ² are selected as the region feature amount, and the 5.0 mm ² holes 22 appearing in the characters 21 are filled. It shows the state that was done.

  The “circular element” is a feature amount for removing noise. For example, the “circular element” is a radius of a circle, which is stored in the initial setting file 15 and is selected by operating the input unit 18. The By performing the erosion process of step S16, as shown in FIG. 6B, even when noise 23 appears in the background portion of the character 21, noise smaller than the size of the circular element selected in advance is removed. The extracted character 21 becomes more probable. FIG. 6B shows a state in which a radius of 2.0 mm is selected as the circular element, and noise with a radius of 1.5 mm appearing in the character 21 is removed. Note that an image with a radius of 3.0 mm is left as it is because it is likely to be a part of a character.

  Next, the process proceeds to step S17, and the connected components of the region are calculated. Thereby, as shown in FIG.6 (c), each character in an image file is extracted.

  Next, the process proceeds to step S18, and a region is selected based on the character shape feature amount. Here, the “character shape feature amount” designates the width and height of the character, is stored in the initial setting file 15, and is selected by operating the input means 18. For example, as shown in FIG. 6D, as the “character shape feature amount”, the minimum value of the width is 5 mm, the maximum value of the width is 10 mm, the minimum value of the height is 15 mm, and the maximum value of the height is 20 mm. Is selected, characters having a width of 3 mm and a height of 10 mm and characters having a width of 5 mm and a height of 10 mm are ignored, and only characters of a predetermined size are extracted.

  Next, after performing the erosion process again using the circular element (step S19) and the calculation of the connected components of the area (step S20), as shown in FIG. A tolerance area with the area is calculated (step S21). Further, a character region is selected based on the above-described shape feature amount (step S22), and the character region is rearranged according to its relative position (step S23). For example, as shown in FIG. 6F, when the character display position is high or low, the positions of the upper left corners of the character areas are compared, and the character areas are arranged so that the height positions of the character areas are aligned. Sort by. This facilitates comparison with the reference character string. In the case where the stamped characters are displayed on the steel material A using a punch or the like, it is difficult to display the characters on the steel material A by aligning the character strings in one line, so this rearrangement process is particularly effective.

  Thereafter, the process proceeds to step S24, character recognition is performed, and the recognized character data is stored in a data storage unit (not shown) in the character recognition unit 14. The processing from step S12 to step S25 is performed until it is determined in step S26 that the binarization threshold loop of step S11 is completed, and until it is determined in step S27 that the loop for each rotation angle of step S10 is completed. Repeatedly, if it is determined in step S26 and step S27 that the loop is completed, the process proceeds to the process (3) shown in FIG.

  The process (3) shown in FIG. 7 is the latter half of the character recognition algorithm. After the processes in FIGS. 5 and 6 are completed, a loop for each arrangement of recognized character areas is started (step S28). Next, the process proceeds to step S29, where the number of recognized characters for each recognized character area, that is, the number of characters in the recognized character area is checked. Thereby, the recognition of the character recognition in consideration of the number of digits of the characters constituting the character string can be performed, and the reliability of the recognized character string can be improved. In this example, the recognition number for each arrangement of the recognized character area is set to “8”, “9”, “10”, “7”, “6”, “5”, “4”, “3”, “ Loop in the order of “2” and “1”. The reason why the recognition number for each arrangement of the recognized character areas is detected in this order is that the number of characters displayed on the steel material A is large in 8 digits, 9 digits, 10 digits, and 7 digits. This process is repeated until the average evaluation rate of the recognized character area reaches the minimum average evaluation rate (steps S30 and S31).

  If it is determined in step S30 that the average evaluation rate of the recognized character area is larger than the minimum average evaluation rate, the process proceeds to step S32, and the check flag for character data is turned ON (step S32). After the check of the number of characters in step S29 is completed, the process proceeds to step S33, and it is determined whether or not the check flag is ON. If the check flag is not ON, the process proceeds to step S34 to determine whether or not the loop for each recognized character area array in step S28 is completed. If the check flag is ON, step S34 is performed. The process proceeds to S35 to determine whether there is data to be extracted. The same applies to the case where it is determined in step S34 that the loop for each recognized character area array has been completed.

  If it is determined in step S35 that there is data to be extracted, the process proceeds to step S36, where reference characters are acquired from the reference character image storage unit 13 and extracted from character data stored in a data storage unit (not shown). The power recognition area is acquired (step S37), and the character recognition (step S38) is compared. If it is determined in step S35 that there is no data to be extracted, the process returns to step S6 in FIG.

  The process (4) shown in FIG. 8 is the latter half of the character recognition algorithm. First, while the recognized character data is looped by the number of characters (step S39), the evaluation rate, that is, the recognized character data is obtained for each character data. And the matching degree of the reference character image are evaluated (step S40). If it is determined in step S40 that the evaluation rate is greater than the minimum evaluation rate stored in the initial setting file 15 and selected by operating the input means 18, a character for each digit is selected from the character recognition area. (Step S41), the character is added to the final character recognition area array (Step S42). If it is determined in step S40 that the evaluation rate is smaller than the minimum evaluation rate, the processing from step S39 is repeated until it is determined in step S43 that the character data loop has ended.

  If it is determined in step S43 that the loop for all character data has been completed, the process proceeds to step S44, where character recognition is performed with the final character area array, and whether or not the check flag is OFF in step S45. Determine. If it is determined in step S45 that the check flag is not OFF, the recognition result is displayed on the display means 19 and stored in a data storage unit in the character recognition unit 14 (not shown) (step S46). If it is determined in step S45 that the check flag is ON, and if it is determined in step S47 that the loop for all image files has not ended, the process returns to step S6 in FIG.

  The character display part B of the steel material A is displayed with a total of 8 digits divided into 3 levels, the first level is 5 digits, the second level is 2 digits, and the third level is 1 digit In this case, when all the characters are recognized, as shown in FIG. 2 (a), the display means 19 displays the first to eighth digit characters in three stages. In addition, as shown in FIG. 2 (b), the mark “?” Indicating that the character was not recognized among the characters of each digit included in each step is indicated by the corresponding step. It is displayed in the digit. By doing so, the reliability of other recognized characters can be increased.

  Since the character recognition device of this example has such a configuration, it is possible to accurately recognize the characters displayed on the steel material A without being affected by red rust or surface irregularities.

  INDUSTRIAL APPLICABILITY The present invention can be used for a character recognition device particularly suitable for recognizing stamped characters displayed on steel materials.

DESCRIPTION OF SYMBOLS 1 Illuminating means 2 Imaging means 3 Image processing means 11 Camera interface 12 Image file part 13 Reference character image storage part 14 Character recognition part 15 Initial setting file 16 Image processing library 17 Recognition result file 18 Input means 19 Display means A Steel material B Character display Part

Claims (4)

An image pickup means for picking up an image of a steel character display section and outputting a color image signal; and a color image signal input from the image pickup means is subjected to necessary digital image processing to recognize a character displayed on the character display section. Image processing means for performing, and display means for displaying the recognition result of the character by the image processing means,
The image processing unit generates a color image of the character display unit from the color image signal, and then decomposes the color image into channel images of a red channel image, a green channel image, and a blue channel image, and the green channel image And a reference character image stored in advance in the image processing means to evaluate the degree of coincidence between these images. If the degree of coincidence between both images is equal to or greater than a predetermined value, the character is recognized. The character is displayed on the display means, and if the degree of coincidence between the two images is less than the predetermined value, the display means displays an unacceptable display. Character recognition device.   2. The character recognition apparatus according to claim 1, wherein the image processing means repeatedly compares the rotation of the green channel image and the reference character image, and evaluates the degree of coincidence between the two images.   For each of a plurality of characters displayed on the steel material, the image processing means compares the green channel image with the reference character image for each character display position, and evaluates the degree of coincidence between these images. The character recognition apparatus according to claim 1, wherein:   The said image processing means performs the filling process of a hole, the erosion process by a circular element, and the rearrangement process by a relative position with respect to the said green channel image as said digital image process, The rearrangement process by a relative position is characterized by the above-mentioned. 4. The character recognition device according to any one of items 3.

Images