JP2017207910A - Character recognition device, character recognition program and character recognition method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique capable of more easily recognizing a numerical value at a counter part.SOLUTION: A character recognition device includes: an image acquisition part 101 for acquiring a picked-up image in which an area including a plurality of characters in parallel is imaged in gradation; a filter processing part 102 for generating a binary image in which the picked-up image is binarized; a character extraction part 104 for extracting a plurality of partial images respectively including a plurality of characters in the picked-up image on the basis of a lengthwise direction projection in which a total brightness is calculated for each ordinate of the binary image and a crosswise direction projection in which a total brightness is calculated for each abscissa of the binary image; and a matching processing part 107 for recognizing characters by matching each of the plurality of partial images with previously stored dictionary data.SELECTED DRAWING: Figure 3

Description

  Embodiments described herein relate generally to a technique for recognizing characters.

  Conventionally, billing work by a company that supplies power, water, gas, and the like is performed by an operator reading a meter indicating the supply amount. Such meter reading work is performed by an operator visually observing a counter unit that indicates a numerical value by a plurality of numerical drums arranged in parallel in the meter. Therefore, human error such as erroneous reading of the numerical value may occur.

  On the other hand, a model image for collation in which numbers from 0 to 9 are arranged vertically is prepared, and image areas of each digit are individually cut out from the image data obtained by imaging the indicated value display section of the current character type instrument, and the cut image is used. There is known a current-type instrument recognition processing device that extracts a numerical part, compares the numerical part with a model image for verification, performs verification, and reads an indicated value (Patent Document 1).

JP 2003-242444 A

  However, according to the above-described technique, it is necessary to manually set the position of each numerical value in the counter unit to be recognized, and there is a problem that the operation is troublesome when the numerical value of the counter unit is recognized. .

  Embodiments of the present invention have been made to solve the above-described problems, and an object of the present invention is to provide a technique that can more easily recognize the value of the counter unit.

  In order to solve the above-described problem, an embodiment of the present invention includes an image acquisition unit that acquires a captured image in which a region including a plurality of parallel characters is captured with gradation, and a binary that is obtained by binarizing the captured image. A filter processing unit that generates a binarized image, a vertical projection that calculates a total luminance value for each ordinate of the binarized image, and a horizontal projection that calculates a total luminance value for each abscissa of the binarized image Based on the above, a character extraction unit that extracts a plurality of partial images including each of a plurality of characters in the captured image, and a character stored by matching each of the plurality of partial images with prestored dictionary data are recognized. A matching processing unit.

It is a figure showing the whole meter reading system composition containing a meter reading device with which a character recognition device concerning an embodiment was built. It is the schematic which shows the structure of a water meter. It is the schematic which shows the structure of the display part of a water meter. It is a block diagram which shows the hardware constitutions of a character recognition apparatus. It is a block diagram which shows the function structure of a character recognition apparatus. It is a flowchart which shows the whole operation | movement of a character recognition apparatus. It is a figure which shows a captured image. It is a flowchart which shows the operation | movement of a filter process. It is a figure which shows the kernel of a Sobel filter. It is a figure which shows an edge emphasis image. It is a figure which shows a binarized image. It is a figure which shows a thinning image. It is a flowchart which shows operation | movement of a rotation correction process. It is a figure which shows the boundary line L and rotation angle (theta). It is a flowchart which shows the operation | movement of a character area extraction process. It is a figure which shows the object area | region in a binarized image. It is a figure which shows the object area | region in a captured image. It is a figure which shows the noise in a binarized image. It is a figure which shows a y direction projection. It is a flowchart which shows operation | movement of a 1st character area extraction process. It is a figure which shows x direction projection. It is a figure which shows a partial captured image. It is a flowchart which shows the operation | movement of a position correction process. It is a flowchart which shows the operation | movement of a parallel character area extraction process. It is a flowchart which shows the operation | movement of a character determination process. It is a figure which shows two area | regions divided | segmented by the diagonal line which connects an upper right corner and a lower left corner. It is a figure which shows two area | regions divided | segmented by the diagonal connecting the upper left corner and the lower right corner. It is a figure which shows two area | regions divided | segmented into right and left. It is a figure which shows the image used for the character determination by a total luminance value. It is a figure which shows the extraction result by a character area extraction process. It is a graph which shows the correlation with the distance between character height and an extraction window. It is a figure which shows the result which represented the binary image which binarized the partial captured image, and the feature-value calculation process result with the image. It is a figure which shows a feature-value pattern. It is a flowchart which shows operation | movement of a matching process. It is a figure which shows the deformation | transformation location of a number. It is a figure which shows the kernel function used for a matching process. It is a figure which shows the difference between 5 and 6 expressed by the feature-value image. It is a flowchart which shows operation | movement of a counter value determination process. It is a figure which shows the counter part which changes from 499 to 500. It is a figure which shows the counter part with insufficient rotation.

  Embodiments of the present invention will be described below with reference to the drawings.

(overall structure)
A meter reading system including a meter reading device incorporating a character recognition device according to an embodiment, a configuration of a water meter that is a character recognition target by the character recognition device, and a configuration of a display unit of the water meter will be described. FIG. 1 is a diagram illustrating an overall configuration of a meter reading system including a meter reading system including a meter reading device according to an embodiment. FIG. 2 is a schematic diagram showing the configuration of the water meter. FIG. 3 is a schematic diagram showing the configuration of the display unit of the water meter.

  As shown in FIG. 1, the meter reading system reads the flow rate of water displayed on the display unit 93 of the water meter 9 and transmits the read flow rate by wireless communication, and receives the transmitted flow rate. Smartphone 2 to be provided.

  As shown in FIG. 2, the water meter 9 includes a main body 90 that measures the flow rate of water, an inlet 91 that allows water to flow into the main body 90, an outlet 92 that allows water to flow out from the main body 90, and a main body A display unit 93 provided on an upper part of the unit 90 and displaying information related to the flow rate of water measured by the main body unit 90. The water meter 9 is installed such that the display unit 93 faces upward when in use.

  As shown in FIG. 3, the display unit 93 includes a display panel 930 configured in a disk shape, a pilot 931 provided on the display panel 930 so as to be visible, an integrated value meter 932 in 1 L units, and an integrated value meter 933 in 10 L units. , And a counter unit 934 that indicates the flow rate by a numerical value in cubic meters. In the display panel 930, at least a region where the counter unit 934 is provided is colored semi-circularly black, the other regions are colored white, and the boundary between these regions is a straight line. The counter unit 934 indicates the flow rate by an integer of 4 digits or 5 digits and a decimal number of 1 digit, and these numerical values are obtained by rotating a numerical drum in which numerical values of 0 to 9 are arranged in an annular shape on the side surfaces. It is changed by. Number drums showing 4-digit or 5-digit integers are colored black and the numbers are colored white, and number drums showing single-digit decimal numbers are colored white and the numbers are red. It shall be colored. It should be noted that the number drum indicating the integer may be any one in which the brightness of the background color and the brightness of the number color are different from each other. These number drums are incorporated in a space formed by the main body 90 and the display board 930 and are visible from the outside through an opening provided in the display board 930. In addition, when the water meter 9 is installed, the surface direction of the display panel 930 is assumed to be in the horizontal direction.

  As shown in FIG. 1, the meter reading device 1 includes a main body unit 10, a character recognition device 11 built in the main body unit 10, a substantially cylindrical imaging hood 12 provided at a lower portion of the main body unit 10, and an imaging. A camera 13 provided in the hood 12, an LED 13 a that irradiates red light to an object to be imaged by the camera 13, an operation unit 14 for operating the meter reading device 1, and various information related to the meter reading device 1. And a display 15 for displaying. The meter reading device 1 reads the integer part of the counter unit 934 in the display unit 93 of the water meter 9. Therefore, the imaging hood 12 is configured to be placed on the main body unit 9, and the camera 13 is configured to be the imaging hood 12. In contrast, it is assumed that substantially the entire display panel 930 can be imaged. In addition, the user of the meter reading device 1 places the meter reading device 1 on the water meter 9 so that the imaging portion of the camera 13 includes the counter unit 934 and the boundary portion of the display panel 930. .

(Configuration of character recognition device)
The hardware configuration and functional configuration of the character recognition device will be described. FIG. 4 is a block diagram illustrating a hardware configuration of the character recognition device. FIG. 5 is a block diagram illustrating a functional configuration of the character recognition device.

  As shown in FIG. 4, the character recognition device 11 includes, as hardware, an MPU (Micro Processing Unit) 110, a memory 111, a storage device 112, a communication unit 113, and an input / output I / F (interface) 114. The MPU 110 and the memory 111 cooperate to execute various functions, and the storage device 112 stores various data used for processing executed by the various functions. The communication unit 113 communicates with the smartphone 2. The input / output I / F 114 performs data input / output with the camera 13, the LED 13 a, the operation unit 14, and the display 15.

  As shown in FIG. 5, the character recognition device 11 includes, as functions executed by the MPU 110 and the memory 111, an image acquisition unit 101, a filter processing unit 102, a rotation correction unit 103, a character extraction unit 104, and a scale adjustment unit 105. , A feature amount calculation unit 106, a matching processing unit 107, a counter value determination unit 108, and a counter value transmission unit 109. The image acquisition unit 101 acquires a captured image captured by the camera 13. The filter processing unit 102 applies various filters to the acquired captured image. The rotation correcting unit 103 rotates the captured image based on the image to which the filter is applied so that the captured display unit 93 is upright. The character extraction unit 104 extracts a partial image including individual numerical values in the counter unit 934 from the captured image. The scale adjustment unit 105 adjusts the scale of the partial image. The feature amount calculation unit 106 calculates a feature amount from the partial image. The matching processing unit 107 recognizes the numerical value in the partial image by matching the feature amount of the partial image with a feature amount dictionary prepared in advance. The counter value determination unit 108 performs determination on the recognized numerical value of the counter unit 934. The counter value transmitting unit 109 transmits the numerical value of the recognized and determined counter unit 934 to the smartphone 2 via the communication unit 113.

(Whole operation of character recognition device)
The overall operation of the character recognition device will be described. FIG. 6 is a flowchart showing the overall operation of the character recognition apparatus. FIG. 7 is a diagram illustrating a captured image.

  As shown in FIG. 6, first, the image acquisition unit 101 acquires a monochrome grayscale image as shown in FIG. 7 as a captured image captured by the camera 13 (S101). Here, since the number of the decimal part in the counter unit 934 in the captured image is colored red on a white background, it is a white rectangular part by the red light irradiated by the LED 13a. In, it is assumed that it is not recognized and treated as a non-numeric area.

  Next, the filter processing unit 102 executes filter processing described later (S102), and the rotation correction unit 103 executes rotation correction processing described later (S103).

  Next, the character extraction unit 104 executes character region extraction processing (described later) on the captured image erected by the rotation correction processing (S104), and the scale adjustment unit 105 extracts each part extracted by the character region extraction processing. The scale of the image is adjusted (S105), the feature amount calculation unit 106 executes a feature amount calculation process described later (S106), and the matching processing unit 107 recognizes a numerical value in each partial image (S107).

  Next, the counter value determination unit 108 performs a counter value determination process described later (S108), and the counter value transmission unit 109 transmits the counter value of the counter unit 934 determined by the counter value determination process to the smartphone 2 (S109). .

(Filter processing)
The filtering process will be described. FIG. 8 is a flowchart showing the operation of the filter process. FIG. 9 is a diagram illustrating a kernel of the Sobel filter. FIG. 10 is a diagram illustrating an edge enhanced image. FIG. 11 is a diagram illustrating a binarized image. FIG. 12 shows a thinned image.

  As shown in FIG. 8, first, the filter processing unit 102 applies a vertical kernel as shown in FIG. 9A to the captured image acquired by the captured image acquisition unit 101, and FIG. 9B. By applying a Sobel filter using a horizontal kernel as shown, an edge-enhanced image in which an edge portion having a large luminance change is emphasized as shown in FIG. 10 is generated (S201).

  Next, the filter processing unit 102 binarizes the edge-enhanced image with a predetermined threshold value to generate a binarized image as shown in FIG. 11 (S202). Further, for the binarized image, Thinning processing is performed to leave the line only by one pixel width, and a thinned image as shown in FIG. 12 is generated (S203).

  As described above, according to the filter processing, a binarized image and a thinned image used for processing to be described later are generated.

(Rotation correction processing)
The operation of the rotation correction process will be described. FIG. 13 is a flowchart showing the operation of the rotation correction process. FIG. 14 is a diagram illustrating the boundary line L and the rotation angle θ.

  As shown in FIG. 13, first, the rotation correction unit 103 detects a boundary portion on the display panel 930 based on the thinned image (S301). Here, the rotation correction unit 103 is colored in the boundary portion of the display panel 930, that is, black by detecting the longest straight line as shown in FIG. A straight portion of a semicircular area is detected.

  Next, the rotation correction unit 103 detects an angle θ between the horizontal line y = 0 and the boundary line L (S302), and rotates the captured image and the binarized image by −θ so that the angle θ becomes zero. (S303). Here, the binarized image is an image generated in step S202 described above. In the following description, it is assumed that the upper left corner of the image is the origin, the x axis is in the horizontal direction rightward, and the y axis is in the vertical direction downward.

  After the rotation of the image, the rotation correction unit 103 determines whether the rotation-corrected captured image is in an inverted state (S304). Here, the inverted state is a state in which the display unit 93 is inverted in the rotation-corrected captured image, that is, a state in which the numbers in the counter unit 934 are inverted up and down although the boundary lines are parallel. The rotation correction unit 103 adds luminance values to 10 pixels on the upper side and 10 pixels on the lower side from the point on the boundary line in the captured image corresponding to the line in which the boundary line L in the thinned image is rotated. When the upper total value is larger than the lower total value, it is determined that the head is inverted.

  In the inverted state (S304, YES), the rotation correcting unit 103 rotates the rotation-corrected captured image and the rotation-corrected binarized image by 180 ° (S305), and ends the rotation correction process.

  On the other hand, when it is not the inverted state (S304, NO), the rotation correction unit 103 ends the rotation correction process.

  As described above, according to the rotation correction process, the user of the meter reading device 1 can read the counter value of the water meter 9 so that the display unit 93 stands upright on the captured image, that is, the number in the counter unit 934. It is not necessary to install the meter reading device 1 so that the vertical direction of the image and the vertical direction of the captured image coincide with each other.

(Character area extraction processing)
The operation of the character area extraction process will be described. FIG. 15 is a flowchart showing the operation of the character area extraction process. FIG. 16 is a diagram illustrating a target area in a binarized image. FIG. 17 is a diagram illustrating a target area in a captured image. FIG. 18 is a diagram illustrating noise in a binarized image. FIG. 19 is a diagram showing y-direction projection.

  As shown in FIG. 15, the character extraction unit 104 uses a captured image and a binarized image rotated so that the display unit 93 is erected by the rotation correction unit 103, as shown in FIGS. 16 and 17. Is extracted (S401). Here, the target area is an area located above the boundary line, having a predetermined distance in the vertical direction, and including the counter unit 934. At this time, the distance between the lower end of the target area and the boundary line is set so that the decimal point is not included in the target area. For example, when the decimal point portion is 10 pixels above the boundary line, an area from a position 60 pixels above the boundary line to a position 10 pixels above the boundary line is set as the target area.

  Next, the character extraction unit 104 removes noise N as shown in FIG. 18 from the binarized image extracted as the target region (hereinafter simply referred to as a binarized image; the same applies to the captured image) (S402). . The noise N is caused by an opening provided in the display panel 930. The character extraction unit 104 determines that white pixels that are continuous by a predetermined width or more in the horizontal direction are noise N, and changes the corresponding portion to a black pixel. To do. Here, the predetermined width may be at least the character width of one digit in the counter unit 934, and is set to 30 pixels, for example. Further, the character extraction unit 104 determines that the white pixel existing in the upper and lower pixels from the white pixel continuous by a predetermined width or more is also determined as noise N and changes it to a black pixel.

After removing the noise N, the character extraction unit 104 calculates the y-direction projection P _y of the binarized image using the following formula (S403). Here, the projection P _y indicates a total luminance value calculated for each y coordinate.

F ₁ is a binarized image, and W ₁ is the width of the binarized image. An example of a projection _{P y} in Figure 19.

Next, the character extracting unit 104 calculates the character height _{H c} based with predetermined thresholds _{T 1} which is set in advance in the projection _{P y} (S404). Here, the character extraction unit 104 searches for a section in the y direction indicated by the two y coordinates such that P _y (y) ≧ T _1, and among the corresponding sections, the maximum section is the character height. Let _Hc . The character height _Hc is used as a range in the y direction of the target area. Thereby, the ratio of the area | region which comprises the character which occupies for a process target area | region can be increased, and the recognition accuracy of a character can be raised by extension.

  Next, the character extraction unit 104 executes a first character area extraction process to be described later (S405), further executes a parallel character area extraction process to be described later (S406), and ends the character area extraction process.

(First character area extraction processing)
The operation of the first character area extraction process will be described. FIG. 20 is a flowchart showing the operation of the first character area extraction process. FIG. 21 is a diagram showing x-direction projection. FIG. 22 is a diagram illustrating a partially captured image.

As shown in FIG. 20, the character extraction unit 104 first calculates the x-direction projection P _x of the binarized image using the following equation (S501). Here, the projection P _x indicates a total luminance value calculated for each x coordinate.

H ₁ is the height of the binarized image. An example of a projection _{P x} Figure 21.

Next, the character extraction unit 104 calculates the center-of-gravity coordinates xg in the x direction of the projection P _x by the following formula (S502).

Next, using the window width Ws preset in the x direction, the character extraction unit 104 provides an extraction window having a width (2 · Ws) in the projection P _x and sets the center of the extraction window in the x direction to xg calculating the sum of the projection P _x within the extraction window while moving between xg + Ws from the left by one pixel to the right from -ws. The total projection P _x in the extraction window to calculate the center coordinates x _c of the extraction window having the maximum (S504).

Then, the character extraction unit 104 corrects the x-direction position of the center coordinates x _c of the extraction window by the position correcting process which will be described later (S505), the image of the region within the extraction window in the captured image as shown in FIG. 22 Extracted as a partially picked-up image (S506), and the character region extraction processing is terminated. FIG. 22 shows a partial image including the first place “5” in the counter value.

Thus, by calculating the center coordinate x _c in the extraction window barycentric coordinates Xg projection P _x in the binarized image as an initial value, compared with the case of setting the center of the extraction window barycentric coordinates Xg, more characters The extraction window can be set first for regions where there is a high probability of being present. Note that this extraction window is moved to the left and right by the parallel character area extraction process and is corrected to the optimum position.

(Position correction processing)
The position correction process will be described. FIG. 23 is a flowchart showing the operation of the position correction process.

As shown in FIG. 23, first, the character extracting unit 104 sets the extraction window having a window width Ws in lateral to the center coordinate x _c in the rotation correction captured image, the rotation correction in this extraction window The captured image is extracted as a partially captured image (S601), and the partially captured image is binarized by a discriminant analysis method (S602).

Next, the character extraction unit 104 calculates the centroid coordinates x _c g in the x direction in the binarized partially captured image by the following formula (S603).

F _2B is a binarized partially captured image, and H ₂ is the height of the partially captured image.

Next, the character extracting unit 104 determines whether the center coordinates _{x c} and the center of gravity coordinates _{x c} g match (S604). Here, the character extraction unit 104 determines that the center coordinate x _c and the center-of-gravity coordinate x _c g match when x _c g−Ws = 0.

When the center coordinates _{x c} and the geometric center coordinates _{x c} g match (S604, YES), the character extracting unit 104 ends the position correction processing.

On the other hand, if the center coordinate x _c and the center-of-gravity coordinate x _c g do not match (S604, NO), the character extraction unit 104 shifts the center coordinate x _c in the x direction and moves the extraction window (S605). The rotation-corrected captured image in the extracted window is extracted as a partially captured image (S601). In the movement of the extraction window, the center coordinates x _c, a value obtained by adding the center coordinate x _c to a value obtained by subtracting the window width Ws from the center of gravity coordinates x _{c g} is substituted.

As described above, according to the position correction process, by changing the position of the extraction window in the x direction so that the center coordinate x _c and the center of gravity coordinate x _c g coincide with each other, the character based on the partially captured image extracted by the extraction window is displayed. The accuracy of extraction can be improved.

(Parallel character area extraction processing)
The operation of the parallel character area extraction process will be described. FIG. 24 is a flowchart showing the operation of the parallel character area extraction process.

As shown in FIG. 24, first, the character extracting unit 104 moves the center coordinate x _c in the extraction window that is set to the binary image and the captured image into predetermined character spacing value to the right direction (S701), A character determination process to be described later is executed (S702), and it is determined whether or not it is determined by this character determination process that there is a character in the extraction window (S703).

If in the extraction window is not determined that there is a character (S703, NO), the character extracting unit 104, the center coordinates x _c of the set extraction window by the first character area extraction processing to the binary image and the captured image characters The character is moved leftward by the interval value (S704), character determination processing is executed (S705), and it is determined whether or not it is determined by this character determination processing that there is a character in the extraction window (S706).

  If it is not determined that there is a character in the extraction window (S706, NO), the character extraction unit 104 ends the parallel character region extraction process.

On the other hand, if it is determined that there is a character in the extraction window (S706, YES), the character extraction unit 104 corrects the position of the center coordinates _{x c} of the extraction window by the above-described position correction processing (S707), extraction window extracting a captured image area of the inner as partial captured image (S 708), again, moving the center coordinate x _c of the extraction window to character spacing values to the left direction. (S704).

Further, in step S703, if it is determined that there is a character in the extraction window (S703, YES), the character extraction unit 104 corrects the position of the center coordinates _{x c} of the extraction window by the above-described position correction processing (S709 ), rotation correction captured image in the extraction window is extracted as a partial captured image (S710), again, moving the center coordinate x _c of the extraction window to character spacing values to the right direction (S701).

  As described above, according to the parallel character area extraction process, the numerical window parallel to the area specified by the first character area extraction process is specified by moving the extraction window determined by the first character area extraction process left and right. can do. Further, in the search in the left and right directions by the parallel character area extraction process, when the character determination process determines that there is no character in the extraction window, the extraction window has an end portion outside the numerical value portion of the counter unit 934. It is considered that it has reached the side and is finished.

(Character determination processing)
The operation of the character determination process will be described. FIG. 25 is a flowchart showing the operation of the character determination process. FIG. 26 is a diagram illustrating two regions divided by a diagonal line connecting the upper right corner and the lower left corner. FIG. 27 is a diagram illustrating two regions divided by a diagonal line connecting the upper left corner and the lower right corner. FIG. 28 is a diagram showing two regions divided into left and right. FIG. 29 is a diagram illustrating an image used for character determination based on the total luminance value. FIG. 30 is a diagram illustrating an extraction result obtained by the character region extraction process.

  As shown in FIG. 25, first, the character extraction unit 104 determines whether the total luminance value in the extraction window in the binarized image is less than a predetermined ratio (S801). Here, specifically, the character extraction unit 104 determines whether or not the number of white pixels is less than 1/16 of the total number of pixels in the extraction window. According to this determination, when there are extremely few white pixels in the extraction window, it is determined that no character exists in the extraction window.

If the total luminance value is not less than the predetermined ratio (S801, NO), the character extraction unit 104 divides the region in the extraction window in the binarized image by a diagonal line from the upper right corner to the lower left corner as shown in FIG. Region A and region B (S802), and it is determined whether or not there is a bias in the total luminance value between these regions (S803). Here, the character extraction unit 104, specifically, when the total luminance value in the region A S _A, the total luminance value in the region B was set to S _B, if S _A is greater than or equal to 5 times the S _B, Alternatively, it is determined that there is a bias in the total luminance value between the region A and the region B when S _B is 5 times or more of S _A.

When there is no luminance value bias between the regions (S803, NO), the character extraction unit 104 divides the region in the extraction window in the binarized image by a diagonal line from the upper left corner to the lower right corner as shown in FIG. Region A and region B (S804), and it is determined whether there is a bias in the total luminance value between these regions (S805). Here, specifically, the character extraction unit 104, when the total luminance value in the region A is S _A and the total luminance value in the region B is S _B , as in the above determination, S _A is S _B. If it is 5 times or more, or determines that the S _B is skewed a total luminance value between the regions a and B when it is more than five times the S _a.

When there is no luminance value bias between the regions (S805, NO), the character extraction unit 104 divides the region in the extraction window in the binarized image into a central vertical line as shown in FIG. B (S806), it is determined whether there is a bias in the total luminance value between these areas (S807). Here, the character extraction unit 104, specifically, when the total luminance value in the region A _{S A,} the total luminance value in the region B was set to _{S B,} if _{S A} is greater than or equal to 10 times the _{S B,} Alternatively, when S _B is 10 times or more of S _A , it is determined that there is a bias in the total luminance value between region A and region B.

  When there is no bias of the total luminance value between the regions (S807, NO), the character extraction unit 104 adds the image to which the Sobel filter is applied and the original image for the region in the extraction window in the captured image, and further determines. A determination partial image binarized by the discrimination method is generated (S808). After the generation, the character extraction unit 104 calculates the total luminance value in the extraction window in the binarized image as shown in FIG. 29A and the total luminance value in the determination partial image as shown in FIG. It is determined whether or not the difference is greater than or equal to a predetermined value set in advance (S809). In the present embodiment, the predetermined value used for this determination is 175.

  If the difference between the total luminance values is not greater than or equal to the predetermined value (S809, NO), the character extraction unit 104 determines that there is a character in the area within the extraction window (S810), and ends the character determination process.

  On the other hand, if the difference between the total luminance values is equal to or greater than the predetermined value (S809, YES), the character extraction unit 104 determines that there is no character in the area in the extraction window (S811), and ends the character determination process.

  In step S807, if there is a bias in the total luminance value between the regions (S807, YES), the character extraction unit 104 determines that there are no characters in the region in the extraction window (S811).

  In step S805, if there is a bias in the total luminance value between the regions (S805, YES), the character extraction unit 104 determines that there are no characters in the region in the extraction window (S811).

  In step S803, if there is a bias in the total luminance value between the regions (S803, YES), the character extraction unit 104 determines that there are no characters in the region in the extraction window (S811).

  In step S801, when the total luminance value is less than the predetermined ratio (S801, YES), the character extraction unit 104 determines that there is no character in the area in the extraction window (S811).

  Thus, according to the character determination process, it is possible to determine whether or not a character exists in the extraction window based on the total luminance value in the extraction window. Further, according to the character extraction process including the first character area extraction process including the position correction process and the parallel character area extraction process including the character determination process and the position correction process, as illustrated in FIG. The position of the extraction window is moved in the order of W6. Among these, W1, W3, W4, and W5 are determined to have characters, W2 is determined to have no characters by the determination in step S809, and W6 is determined to have no characters by the determination in step S803. The partial captured images extracted by the extraction windows W1, W3, W4, and W5 are associated with digit values indicating the number of digits. It has come to increase.

(Scale adjustment processing)
The scale process will be described. FIG. 31 is a graph showing the correlation between the character height and the distance between the extraction windows.

  Depending on the distance between the camera 13 and the display unit 93, the difference in the type of the water meter 9, and the like, there are situations where the character size in the captured image differs from the character size used when creating the feature dictionary. On the other hand, the scale adjustment unit 105 adjusts the scale of the partially captured image by the scale adjustment process. Here, the scale factor for scale adjustment is determined by an approximate line obtained from the correlation between the distance between the characters measured in advance as shown in FIG. 31 and the character height. The straight line is y = 0.9709x−5.3309.

Scaling unit 105, first, in a plurality of extraction window that is determined as a character, by calculating the distance between the center coordinate x _c of the extraction window adjacent each substitutes the average value of the calculated distance to the approximate line Calculate the text height. Furthermore, the scale adjustment unit 105 adjusts the scale of the partially captured image based on the calculated character height.

(Feature amount calculation process)
The feature amount calculation process will be described. FIG. 32 is a diagram illustrating a binary image obtained by binarizing the partially captured image and a result of representing the feature amount calculation processing result as an image. FIG. 33 is a diagram illustrating a feature amount pattern.

  The feature amount calculation unit 106 first binarizes the partially captured image as shown in FIG. 32A by, for example, a differential histogram method, and calculates the feature amount as shown in FIG. 32B. To do. Here, the feature amount calculation unit 106 divides the binarized image into three parts in the x direction, the left region, the middle region, and the right region, and based on the luminance value included in each region for each y coordinate (row). Each area is set to 0 or 1. In the present embodiment, 1 is set when the white pixel continues for 3 pixels or more, and 0 is set otherwise. In this way, a feature amount set calculated for each row is created. This is hereinafter referred to as a feature image. In this feature amount image, each y coordinate (row) is represented by 3 bits (values of 0 to 7) in FIG. Thus, by representing the partially captured image as a set of feature amounts, robustness in the matching process is improved, and the partially captured image can be suitable for the matching process described later.

(Matching process)
The matching process will be described. FIG. 34 is a flowchart showing the operation of the matching process. FIG. 35 is a diagram illustrating a deformed portion of a number. FIG. 36 is a diagram showing a kernel function used for the matching process. FIG. 37 is a diagram illustrating the difference between the feature quantities 5 and 6. In FIG.

  As shown in FIG. 34, first, the matching processing unit 107 selects an unselected feature amount image (S901), and selects an unselected feature amount dictionary (S902).

  Here, the feature amount dictionary will be described. The feature amount dictionary is a set of 120 images obtained by capturing the entire circumference by shifting the side surface of the cylindrical number drum by 3 ° in the circumferential direction and calculating the respective feature amounts. As shown in FIG. 35, the character of the counter unit 934 is displayed on the side surface of the number drum, and the cross-sectional shape of this side surface draws an arc, so that the deformation of the character increases as it moves up and down from the center of the number drum. In addition, numerical values of 0 to 9 are arranged on the side surfaces of the numeric drum, and for example, two numerical values arranged adjacent to each other can be visually recognized during the change of the counter value. Therefore, by using not only an image obtained by capturing individual numbers in this way but also an image that considers the middle of a change, matching with higher accuracy can be performed. Each of these 120 feature quantity dictionaries is associated with a value of 0.0 to 9.9. In this value, the integer part indicates a number displayed as a counter value, and the decimal part is Indicates the degree of change to the number adjacent to the rotation direction of the number drum. For example, 9.5 is set in the feature amount dictionary in which a part of 9 is imaged upward and a part of 0 is imaged in the middle while changing from 9 to 0. The value associated with the feature dictionary is used for the counter value determination process.

  Next, the matching processing unit 107 calculates the similarity R between the selected feature image and the feature dictionary using the following formula (S903).

I (h) represents the feature amount of the feature amount image in the h row, and T (h) represents the feature amount of the feature amount dictionary in the h row. The Telson product symbol indicates an operation for determining a kernel value set for a combination of feature amounts, and a kernel function described later is used. Also,

According to this process, the target to be matched with the h row is selected as the matching result with the maximum calculated value in the total 2 · Δh + 1 rows of the feature amount dictionary h−Δh row to h + Δh row. Thereby, even when a scale adjustment error occurs due to the scale adjustment process, the influence of this error on matching can be reduced.

  As shown in FIG. 36, the kernel function used to calculate the similarity R is obtained by individually setting a weight for a combination of feature amounts, and a small value for a combination of feature amounts to be distinguished. Is set. For example, the difference between 5 and 6 expressed by the feature amount image as shown in FIG. 37 is only a part, but here, the feature amount [001] = 1 and the feature amount [101] = 5. By setting a small value for, the discrimination performance of 5 and 6 in the matching process can be improved.

  After calculating the similarity R, the matching processing unit 107 determines whether there is an unselected feature amount dictionary (S904).

  When there is no unselected feature quantity dictionary (S904, NO), the matching processing unit 107 selects a feature quantity dictionary having the maximum calculated similarity R (S905), and is associated with this feature quantity dictionary. A value is acquired (S906), and it is determined whether there is an unselected feature amount image (S907).

  If there is no unselected feature amount image (S907, NO), the matching process is terminated.

  On the other hand, when there is an unselected feature value image (S907, YES), the matching processing unit 107 selects an unselected feature value image again (S901).

  If there is an unselected feature value dictionary in step S904 (S904, YES), the matching processing unit 107 selects an unselected feature value dictionary again (S902).

(Counter value judgment processing)
The counter value determination process will be described. FIG. 38 is a flowchart showing the operation of the counter value determination process. FIG. 39 is a diagram illustrating a counter unit that changes from 499 to 500.

  As shown in FIG. 38, first, the counter value determination unit 109 assigns 0 to a variable n with the least significant digit being 0 (S1001), where n indicates the number of digits of the counter value and is set in advance. It is determined whether or not it is smaller than the number of digits (4 in the present embodiment) (S1002).

  When n is smaller than the set number of digits (S1002, YES), the acquired value acquired by the matching process for n digits (the value associated with the feature dictionary having the maximum similarity R) is 9 + Δn to 10−Δn. It is determined whether it is within a section (S1003). Here, Δn is a parameter that is set in advance and is a parameter that adjusts the determination of whether a carry has occurred. In this embodiment, it is assumed that Δn is set to 0.3.

  When the acquired value is within the interval from 9 + Δn to 10−Δn (S1003, YES), the counter value determination unit 109 adds n digits to the determination pending digit that is a digit to be determined later (S1004), and n Is incremented (S1005), and it is determined again whether n is smaller than the set number of digits (S1002).

  On the other hand, when the acquired value is outside the interval from 9 + Δn to 10−Δn (S1003, NO), the counter value determining unit 109 determines whether or not the decimal part of the acquired value of the nth digit is (1−Δn) or less. Is determined (S1006).

  When the fractional part of the acquired value for n digits is equal to or less than (1-Δn) (S1006, YES), the counter value determining unit 109 sets the integer part of the acquired value for the nth digit as the reading value for the nth digit ( S1007), 0 is substituted into the variable m (S1008), and it is determined whether m is smaller than the maximum determination hold digit which is the maximum among the determination hold digits (S1009).

  When m is smaller than the maximum determination pending digit (S1009, YES), the counter value determination unit 109 determines whether the m digit is a determination pending digit (S1010).

  When the m digit is the determination pending digit (S1010, YES), the m digit reading value is set to 9, the m digit is released from the determination pending digit (S1011), m is incremented (S1012), and m is determined to be the maximum again. It is determined whether it is smaller than the reserved digit (S1009). Here, the read value is a value that is finally read as a counter value of the counter unit 932.

  On the other hand, when the m digit is not the determination pending digit (S1010, NO), the counter value determination unit 109 increments m (S1012).

  In step S1009, when m is equal to or greater than the maximum determination hold digit (S1009, NO), the counter value determination unit 109 increments n (S1005).

  In step S1006, when the fractional part of the acquired value for n digits is larger than (1−Δn) (S1006, NO), the counter value determining unit 109 sets the integer part + 1 of the acquired value for the nth digit to n digits. The read value of the eye (S1013), 0 is substituted for m (S1014), and it is determined whether m is smaller than the maximum determination hold digit (S1015).

  When m is smaller than the maximum determination target hold digit (S1015, YES), the counter value determination unit 109 determines whether the m digit is a determination hold digit (S1016).

  When the m digit is the determination pending digit (S1016, YES), the counter value determining unit 109 sets the m digit reading value to 0, cancels the m digit from the determination pending digit (S1017), and increments m (S1018). Then, it is determined again whether m is smaller than the maximum determination hold digit (S1015).

  On the other hand, when the m digit is not the determination pending digit (S1016, NO), the counter value determination unit 109 increments m (S1018).

  In step S1015, when m is equal to or greater than the maximum determination hold digit (S1015, NO), the counter value determination unit 109 increments n (S1005).

  In step S1002, if n is not smaller than the set number of digits (S1002, NO), the counter value determination unit 109 ends the counter value determination process.

  As described above, according to the counter value determination processing, as shown in FIG. 39, the digit (between 0 and 9) of the captured image when the counter value changes from [0499] to [0500] ( By determining whether or not the digit higher than the determination target digit) is changing to a larger number, the number of the determination target digit can be read correctly. In addition, as can be seen from FIG. 40 showing the counter portion that is not sufficiently rotated, there is a water meter in which the center of gravity in the vertical direction of the number is greatly deviated from the center position even if the number is not in the middle of the change. It is desirable to set the threshold value Δn of the decimal part at about 0.3.

  In the present embodiment, the character recognition program is described as being installed in advance in the above-described character recognition device. However, the character recognition program in the present invention includes one stored in a storage medium. Here, the storage medium is a medium that is removable from the defect inspection apparatus, such as a magnetic tape, a magnetic disk (hard disk drive, etc.), an optical disk (CD-ROM, DVD disk, etc.), a magneto-optical disk (MO, etc.), and a flash memory. Further, it refers to all media that can be read and executed by a computer as the character recognition device described above, such as media that can be transmitted via a network.

  Although an embodiment of the invention has been described, this embodiment is presented as an example and is not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

11 character recognition device, 101 image acquisition unit, 102 filter processing unit, 104 character extraction unit, 107 matching processing unit

Claims (8)

An image acquisition unit that acquires a captured image in which a region including a plurality of parallel characters is captured in gradation;
A filter processing unit that generates a binarized image obtained by binarizing the captured image;
Based on a vertical projection in which a total luminance value is calculated for each ordinate of the binarized image and a horizontal projection in which a total luminance value is calculated for each abscissa of the binarized image, a plurality of images in the captured image A character extraction unit that extracts a plurality of partial images including each of the characters;
A character recognition device comprising: a matching processing unit that recognizes a character by matching each of the plurality of partial images with dictionary data stored in advance.   The character extraction unit determines a horizontal position of a first extraction window having a vertical range based on the vertical projection and a predetermined horizontal width based on a center of gravity of the horizontal projection, and the first extraction window is determined as the first extraction window. The character recognition apparatus according to claim 1, wherein the character recognition apparatus is set to a captured image, and the image in the first extraction window is extracted as a partial image.   The character extraction unit has a vertical range and a horizontal width equivalent to those of the first extraction window, and images the second extraction window that is sequentially moved in units of a predetermined distance in the horizontal direction with respect to the first extraction window. The character recognition apparatus according to claim 2, wherein the character recognition device is set to an image, and the image in the second extraction window is extracted as a partial image.   The character extraction unit sequentially moves the second extraction window in one horizontal direction with respect to the first extraction window, and determines whether or not the region image in the second extraction window includes characters. 4. The character recognition device according to claim 3, wherein the character recognition device makes a determination based on a luminance value, and sequentially moves the region image in a laterally opposite direction with respect to the first extraction window when the region image does not include a character. A feature amount calculation unit that calculates a feature amount of each of the plurality of partial images and generates a feature amount set that is a set of feature amounts;
The said matching process part recognizes a character by calculating the similarity degree between the said dictionary data which is a set of the feature-values memorize | stored previously, and the said feature-value set. The character recognition apparatus as described in any one of.   The matching processing unit calculates the similarity based on a weight set in advance for a combination of a feature amount included in the dictionary data and a feature amount included in the feature amount set. The character recognition device according to claim 5. Acquire a captured image in which a region including a plurality of characters in parallel is captured with gradation,
Generating a binarized image obtained by binarizing the captured image;
Based on a vertical projection in which a total luminance value is calculated for each ordinate of the binarized image and a horizontal projection in which a total luminance value is calculated for each abscissa of the binarized image, a plurality of images in the captured image Extract multiple partial images containing each of the characters
A character recognition program for causing a computer to execute processing for recognizing a character by matching each of the plurality of partial images with dictionary data stored in advance. Computer
Acquire a captured image in which a region including a plurality of characters in parallel is captured with gradation,
Generating a binarized image obtained by binarizing the captured image;
Based on a vertical projection in which a total luminance value is calculated for each ordinate of the binarized image and a horizontal projection in which a total luminance value is calculated for each abscissa of the binarized image, a plurality of images in the captured image Extract multiple partial images containing each of the characters
A character recognition method for recognizing a character by matching each of the plurality of partial images with dictionary data stored in advance.