JP5105440B2 - Information processing apparatus, character recognition method, character recognition program, and computer-readable recording medium on which character recognition program is recorded - Google Patents

Description

  The present invention relates to an information processing apparatus, a character recognition method, a character recognition program, and a computer-readable recording medium on which a character recognition program is recorded, and in particular, an information processing apparatus for recognizing a character surrounded by an outer frame. The present invention relates to a character recognition method, a character recognition program, and a computer-readable recording medium on which the character recognition program is recorded.

  In recent years, character readers such as optical character readers (hereinafter referred to as “OCR”) have been desired to support various reading objects such as business cards and magazines, and the number of characters and symbols that must be recognized has increased. ing. In OCR, a recognizable character means a character pattern registered in advance in a stored dictionary. Characters other than the character patterns registered in the dictionary cannot be recognized.

  In particular, characters with squares (characters surrounded by a square frame) and characters with circles (characters surrounded by a round frame) can be freely combined with symbols such as circle (○) or square (□) and kanji. A large number of character patterns can be created. For example, “intersection” with a square, “hour” with a circle, and the like. These characters are often found in magazines and business texts. For example, “community” is used to represent item information of a store or company such as transportation means and “time” is business hours.

Patent Document 1 discloses a method of cutting out a character pattern in a circle and recognizing the cut out character. As described in Patent Document 1, Hough transform (Non-Patent Document 1), which is a known technique, is mainly used as means for detecting a circle in an image. In Patent Document 1, as a method of cutting out a character pattern of a circle character, a circle is extracted by Hough transformation, a black pixel outside a virtual circle having a slightly smaller radius than the inside of the circle pattern is punched, and a round frame is formed. A method of removal is described.
Japanese Patent Laid-Open No. 6-1111070 Satoshi Ishidate, “Detection of straight lines and circles from images by Hough transform” [online], August 30, 2005, CodeZine, [October 15, 2008 search], Internet <URL: http: // codezine. jp / a / article.aspx? aid = 153,>

  In order to recognize a character surrounded by an outer frame such as a circle or square by OCR, it is necessary to register a character pattern of a character (synthetic character) with an outer frame in the dictionary in advance. However, the outer frame and the characters can be created in any combination. For this reason, it is troublesome to register all such composite characters in a comprehensive manner, and even if they are registered, the amount of processing for comparing the input character pattern and the dictionary character pattern increases.

  Further, according to the method of Patent Document 1, it is described that when a plurality of circles are found by the Hough transform, a circle that does not share the center of the outermost circle among the circles in an inclusive relationship is described. Therefore, if the inner circle is distorted (for example, a circle whose black pixel distribution is not uniform on the right and left sides), the outer circle can be extracted, but the inner circle cannot be extracted. It cannot be removed.

  The present invention has been made in order to solve the above-described problems, and its purpose is to achieve accuracy without registering a character pattern of a composite character with an outer frame such as a circle or square in a dictionary. It is an object to provide an information processing apparatus, a character recognition method, a character recognition program, and a computer-readable recording medium on which a character recognition program is recorded.

An information processing apparatus according to an aspect of the present invention is an information processing apparatus for recognizing a character surrounded by an outer frame from a character string image, and cuts out a rectangular image corresponding to one character from the character string image. Shape determination for determining whether or not the character shape in the rectangular image matches a predetermined type of outer frame shape based on the character cutout means and the distribution of black and white pixels in the cutout rectangular image The width of the outer frame to be removed is determined based on the distribution of black pixels within the range corresponding to the outer frame shape determined to match , And a recognition unit for recognizing the character pattern of the image processed by the frame removal unit. The frame removal unit removes the outer frame of the character in the rectangular image.

Preferably, the frame removing means, a plurality of scanning in a first direction parallel to one side of the rectangular image, the number of consecutive black pixels from the end portion side of the rectangular image is calculated for each scan, it is determined matching the Depending on the outer frame shape, the average value of the number of black pixels calculated in all or part of the plurality of scans is calculated, and the black pixels are inverted to white pixels by the calculated average number of black pixels. To remove the outer frame.

Preferably, the outer frame shape of the predetermined type includes a round shape, and the frame removing unit determines the first direction of the rectangular image when the shape determining unit determines that the character shape in the rectangular image is a round shape. Alternatively, by scanning in the second direction orthogonal to the first direction, the means for calculating the number of continuous black pixels from the ends of the two opposing sides and the average value calculated from the ends of the two opposing sides Means for removing the outer frame by inverting black pixels to white pixels by the number of black pixels.

  Preferably, the outer frame shape of the predetermined type includes a quadrangle shape, and a specific character storage unit for storing in advance a character pattern inside a character whose outer side is a square frame as a specific character pattern, and a rectangular shape by the shape determination unit And a matching unit for matching the character pattern of the image after the processing by the frame removing unit with the specific character pattern when the character shape in the image is determined to be a square shape, If it is determined that the character pattern matches the specific character pattern, the character recognition of the original character pattern cut out by the character cutout means and having the outer frame remaining is performed.

  Preferably, the recognizing unit performs character recognition of the character pattern after the processing by the frame removing unit when the matching unit determines that it does not match the specific character pattern.

  Preferably, the storage unit further includes a storage unit, and the storage unit holds both the image after the processing by the frame removal unit and the rectangular image before the outer frame is removed when it is determined that the shape determination unit matches. To do.

  Preferably, the outer frame shape of the predetermined type includes a rectangular shape, and a character pattern in the rectangular image by the shape determination unit and a specific character storage unit for storing in advance a character pattern that is a rectangular frame as a specific character pattern. When it is determined that the shape is a quadrangular shape, a matching unit is provided for matching the character pattern of the rectangular image with the specific character pattern, and the frame removing unit determines that the matching unit does not match the specific character pattern. Only when it is done, the outer frame of the character in the rectangular image is removed.

Preferably, the frame removing unit scans the first direction of the rectangular image and the second direction orthogonal to the first direction when the shape determining unit determines that the character shape in the rectangular image is a square shape. By means of the above, means for calculating the number of continuous black pixels from the ends of the four sides of the rectangular image, and from each of the ends of the four sides, the black pixels are inverted into white pixels by the calculated average number of black pixels. And means for removing the outer frame.

  Preferably, the character surrounded by the outer frame includes any one of kanji, alphabetic characters, hiragana, numbers, and katakana within the frame.

A character recognition method according to another aspect of the present invention is a method for recognizing a character surrounded by an outer frame from a character string image, and a step of cutting out a rectangular image corresponding to one character from the character string image Determining whether the character shape in the rectangular image matches a predetermined type of outer frame shape based on the distribution of black and white pixels in the cut out rectangular image; and the character shape in the rectangular image Is determined to match the outer frame shape of the predetermined type, the width of the outer frame to be removed is determined based on the distribution of black pixels within the range corresponding to the outer frame shape determined to match, A step of removing the outer frame of the character in the rectangular image based on the width, and a step of performing character recognition on the character pattern of the image after the outer frame is removed.

A character recognition program according to still another aspect of the present invention is a program for recognizing a character surrounded by an outer frame from a character string image, and cuts out a rectangular image corresponding to one character from the character string image. Determining whether or not the character shape in the rectangular image matches a predetermined type of outer frame shape based on the distribution of black and white pixels in the cut out rectangular image; and the character in the rectangular image When it is determined that the shape matches the outer frame shape of a predetermined type, the width of the outer frame to be removed is determined based on the distribution of black pixels within the range corresponding to the outer frame shape determined to match , and determined. Based on the determined width, the computer is caused to execute a step of removing the outer frame of the character in the rectangular image and a step of performing character recognition on the character pattern of the image after the outer frame is removed.

  A computer-readable recording medium according to still another aspect of the present invention records the above-described character recognition program.

  According to the present invention, it is possible to accurately recognize a character surrounded by an outer frame without registering a character pattern of a composite character with an outer frame in a dictionary.

  Embodiments of the present invention will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals and description thereof will not be repeated.

[Embodiment 1]
<About configuration>
First, the configuration of the information processing apparatus in the present embodiment will be described.

  FIG. 1 is a functional block diagram showing a functional configuration of information processing apparatus 1 according to Embodiment 1 of the present invention.

  With reference to FIG. 1, the information processing apparatus 1 is connected to an input device 2 and an output device 13. The input device 2 reads an image from a document such as a magazine and sends image data to the information processing device 1. The output device 13 outputs and displays the character recognition result (character string) of the image in the information processing device 1.

  The information processing device 1 controls the entire input / output control unit 14 for executing data input / output control processing between the input device 2 and the output device 13 and each functional block included in the information processing device 1. The control unit 3, the binarization processing unit 5 for converting the image data sent from the input device 2 into a binary image, and the binarized image from the line-by-line image (hereinafter “character string line”). A character string row cutout unit 6 for cutting out an image), a character cutout unit 7 for cutting out a rectangular image corresponding to one character from the cut out character string row image, and characters in the cut out rectangular image A character shape determination unit 8 for determining whether or not the shape matches the outer frame shape of a predetermined type, and an outer frame of the character in the rectangular image based on the distribution of black pixels when determined to match Frame removing unit 9 for removing the outer frame and the outer frame It includes a matching unit 10 for matching the removed character in advance 1 character dictionary 11 specific character pattern registered in the, and a character recognition processing section 12 for performing character recognition using the character recognition dictionary 4. In addition, the information processing apparatus 1 includes a memory 15 for holding information.

  The character recognition dictionary 4 is a dictionary database that exists conventionally, and stores a plurality of character patterns. In the present embodiment, the characters stored in the character recognition dictionary 4 include at least one of kanji, English characters, hiragana, numbers, and katakana. The character to be recognized is not limited to such a character type, and may be, for example, a Hangul character.

  The one-character dictionary 11 stores in advance a character pattern inside a character whose outer side is a square frame as a specific character pattern. For example, if the character type is a Kanji character, the “character having a square frame on the outside” represents, for example, a Kanji character with “Kunigae Mae” as a radical, that is, “Country”, “Fix”, and the like. In such an example, the specific character pattern represents a pattern of characters “ball” and “old” inside a square frame (Kunigae). In this example, “ball” and “old” inside the square frame are also shown as characters, but the inside of the square frame may be a symbol (a combination of lines and points) that does not make sense. Shall.

  The frame removal unit 9 calculates the number of continuous black pixels from the edge side of the rectangular image, and removes the outer frame by inverting the black pixels to white pixels by the calculated number of black pixels. When the outer frame is rectangular, the number of black pixels continuous from the ends of the four sides of the rectangular image is calculated by scanning the rectangular image in the vertical and horizontal directions. Then, the outer frame is removed by inverting black pixels to white pixels by the calculated number of black pixels from each end of the four sides. A specific example of the removal of the square frame will be described later.

  FIG. 2 is a hardware block diagram of the information processing apparatus 1 according to Embodiment 1 of the present invention.

  Referring to FIG. 2, as information processing apparatus 1, a PC (Personal Computer) may be typically employed. The information processing apparatus 1 includes a CPU (Central Processing Unit) 210 for executing various arithmetic processes, and a hard disk unit (HDD: Hard Disk Drive) for storing programs executed by the CPU 210 and various information in a nonvolatile manner. 212, a memory unit 214 for temporarily storing various data, an operation unit 211 for receiving input of instructions from the user, and an optical medium such as a CD-ROM (Compact Disk-Read Only Memory) 215a Data transmission / reception between the image reading device 221 and the display device 222, a CD-ROM drive 215 for reading out the read data and programs, a communication interface unit 217 for exchanging data with external devices, Input / output I / F (interface) 216. These units exchange data with each other via the internal bus 218.

  The image reading device 221 corresponds to the input device 2 shown in FIG. 1 and may be configured by a scanner or a camera, for example. The display device 222 corresponds to the output device 13 shown in FIG. 1 and may be composed of, for example, a liquid crystal.

The input / output control unit 14 illustrated in FIG. 1 corresponds to the input / output I / F 216.
The control unit 3, the binarization processing unit 5, the character string row cutout unit 6, the character cutout unit 7, the character shape determination unit 8, the frame removal unit 9, the matching unit 10, and the character recognition processing unit 12 illustrated in FIG. This is a function realized by the CPU 210 executing software. Note that at least one of these may be realized by hardware.

  The memory 15 illustrated in FIG. 1 corresponds to the HDD 212 and the memory unit 214, for example. The character recognition dictionary 4 and the one-character dictionary 11 shown in FIG. Note that the memory 15, the character recognition dictionary 4, and the one-character dictionary 11 may correspond to an external removable memory (for example, CD-ROM 215a).

<About operation>
Next, the operation of the information processing apparatus 1 according to Embodiment 1 of the present invention will be specifically described. In the present embodiment, the information processing apparatus 1 will be described as performing only recognition processing of a character with a square (a character surrounded by a square frame).

  FIG. 3 is a flowchart showing character recognition processing executed by the information processing apparatus 1 according to Embodiment 1 of the present invention. The processing shown in the flowchart of FIG. 3 is stored in advance in the HDD 212 as a program, and the function of character recognition processing is realized by the CPU 210 reading and executing this program.

  Referring to FIG. 3, binarization processing unit 5 binarizes an image of a document on which a character string is described (hereinafter referred to as “character string image”) (step S201). Specifically, the image data read by the input device 2 is converted into a binary image represented by binary values of 0 and 1 (white and black).

  There are many known methods for binarization. In the present embodiment, an average value of all pixel values of an image is used as a threshold value for assigning each pixel of image data to 0 and 1. Note that any other method may be used as the binarization method.

  Assume that a character string image (input original) 301 as shown in FIG. In that case, when the character string image 301 is binarized, an image 302 as shown in FIG. 5 is obtained.

  In the present embodiment, the character string image is described as an image read by the input device 2. However, the character string image is not limited and may be an image transmitted from the outside.

  Next, the character string line cutout unit 6 cuts out the binarized image 302 for each line (step S202). That is, the character string line image is cut out. There are many known methods for cutting out character string lines. In the present embodiment, for example, a method of cutting out character string rows from the distribution of blank portions (white pixels) and character portions (black pixels) in the image 302 is used. Any other method may be used.

  FIG. 6 shows an example in which a character string row is cut out from the binarized image 302 shown in FIG. Referring to FIG. 6, four character string rows (images) P <b> 701 to P <b> 704 are cut out from the binarized image 302.

  Next, the character cutout unit 7 selects one of the cut out character string rows, and cuts out the characters one by one from the character string row (step S203). That is, a rectangular image corresponding to one character is cut out from the character string row image as a “character rectangular image”. A method for cutting out a character rectangular image may also be realized by a known technique. In the present embodiment, when a character rectangular image is cut out, it is a condition that the outermost black pixel on the upper, lower, left, and right sides of the black pixel of each character image is cut out at the maximum coordinate.

  It should be noted that a composite character surrounded by a square frame is recognized as one character. Therefore, even if it is a composite character in which two characters are surrounded by a square frame, it is cut out as a rectangular image corresponding to one character.

  In the above example, for example, first, the character string row P701 is selected from the character string rows P701 to P704, and a process of cutting out a character rectangular image from the character string row P701 is performed.

The subsequent processing will be described using the character string row P703 shown in FIG. 6 as an example.
FIG. 7 is a diagram showing a character rectangular image cut out from the character string row P703 shown in FIG.

  Next, the character shape determination unit 8 selects one extracted character rectangular image, and determines whether or not the shape of the image is a square (step S204). That is, it is determined whether or not the character shape is a square from the distribution of black and white pixels in the character rectangular image (hereinafter, such determination is abbreviated as “square determination”). Although the square determination may be performed by a known method, in the present embodiment, for example, the following method is used.

  The square determination will be described using the character rectangular image P801 in the character string row P703 shown in FIG. 7 as an example. The character rectangular image P801 represents an image of a character with a square frame (that is, a combined character of “field” and the square frame) in the Chinese character “place”.

  FIG. 8 is a flowchart showing the flow of the square determination process. FIG. 9 is a diagram for explaining a square determination method.

  As shown in FIG. 9, in the present embodiment, the upper left corner of the character rectangular image P801 is considered as the origin, the horizontal direction (horizontal direction) is the X direction, and the vertical direction (vertical direction) is the Y direction.

  Referring to FIG. 8, first, scanning is performed in the X direction from the origin of character rectangular image P801, and the number of black pixels is counted for each line (step S301). As shown in FIG. 9 (1), the count number of black pixels obtained by scanning in the X direction is stored from Y1 to Yb (b is an arbitrary integer). That is, the number of black pixels obtained by scanning in the X direction is temporarily stored in the memory 15 for each line from the line Y1 to the line Yb.

  Further, similar processing is performed in the Y direction (step S302). That is, scanning is performed in the Y direction from the origin of the character rectangular image P801, and the number of black pixels is counted for each line. As shown in FIG. 9 (1), the count number of black pixels obtained by scanning in the Y direction is stored from X1 to Xb (b is an arbitrary integer). In this case, the number of black pixels obtained by scanning in the Y direction is temporarily stored in the memory 15 for each line from the line X1 to the line Xb.

  Here, in FIG. 9 (1), scanning in the X and Y directions is performed from the end to the end of the image. However, in the case of a square character, the circumscribed rectangle is filled with black pixels, so that scanning is performed by a certain width. For example, it can be determined whether the shape is a square. That is, as shown in FIG. 9B, scanning in the X direction may be performed only up to, for example, 1/8 in the Y direction from the origin of the character rectangular image P801. That is, scanning in the X direction may be performed only for the lines Y1 to Yb in FIG. In this case, similarly, from the lower left end of the character rectangular image P801, for example, up to 1/8 may be performed in the Y direction (scanning in the X direction is performed only from the line Y′1 to the line Y′b).

  Further, the scanning in the Y direction is performed in the same manner. That is, it is performed only up to, for example, 1/8 in the X direction from the origin of the character rectangular image P801 (only the lines X1 to Xb are scanned in the Y direction). Similarly, from the upper right end of the character rectangular image P801, for example, only 1/8 is performed in the X direction (only the lines X'1 to X'b are scanned in the Y direction). Here, the predetermined range to be scanned is set to 1/8. However, the range is not limited, and the number of denominators may be arbitrarily selected and optimal.

In the following description, it is assumed that only a part of the range is scanned as shown in FIG.
The character shape determination unit 8 determines whether or not the character rectangular image P801 has a quadrangular shape from the black pixel distribution obtained by such scanning processing (step S205). If it is determined that the shape is a rectangle (YES in step S205), the process proceeds to step S206. On the other hand, if it is determined that the shape is not a quadrangular shape (NO in step S205), the original character pattern is subjected to character recognition processing (step S210), and the process proceeds to step S211. The character recognition process can be realized by a known method. The character recognition process is an almost established technique, and various character recognition methods have been proposed. Here, OCR (Optical Character Recognition) is used, but any other method may be used.

  If the character rectangular image P801 is rectangular, the number of pixels (count number) close to the number of horizontal pixels of the character rectangular image P801 in black pixels in Y1 to Yb and Y′1 to Y′b in FIG. There should be something stored. Further, there are black pixels in X1 to Xb and X′1 to X′b in FIG. 9B in which the number of pixels (count number) close to the number of vertical pixels of the character rectangular image is stored. It should be.

  Assume that “Y1, Y′1, X1, X′1” stores the number of pixels close to the number of pixels of each width. If these four black pixel count numbers satisfy a certain value, it can be determined that the character pattern in the character rectangular image P801 has a square shape. As this fixed value, if the value is in the X direction, for example, 80% of the number of horizontal pixels (total number of pixels in the X direction) of the character rectangular image P801, and if the value is in the Y direction, the vertical value of the character rectangular image P801. For example, 80% of the number of pixels (the total number of pixels in the Y direction) may be used. This constant value may be a variable value depending on the configuration of the information processing apparatus 1.

  However, when the black pixels in the middle line between Y1 and Yb shown in FIG. 9 (2) are close to the number of horizontal pixels, it is preferable not to determine that they are squares. The same applies to Y′1 and Y′b, X1 and Xb, and X′1 and X′b. If any one of the four sets satisfies the above condition, the square is not determined.

  Here, assuming that it is determined that the character rectangular image P801 has a quadrangular shape, the processing after step S206 will be described.

  In step S <b> 206, the frame removal unit 9 executes a process of removing a square frame in the character rectangular image P <b> 801 (hereinafter also referred to as “square frame removal process”).

  FIG. 10 is a flowchart showing the square frame removal process. FIG. 11 is a diagram for explaining the square frame removal process. As shown in FIG. 11A, the upper left corner of the character rectangular image P801 is the origin, the horizontal direction is the X direction, and the vertical direction is the Y direction.

  First, the frame removal unit 9 starts scanning in the X direction from the origin, and counts the number of black pixels that appear continuously from the black pixels that appear first (step S401). If a white pixel appears, the scanning of that line ends. The number of pixels obtained at this time is stored in Y1. Similarly, scanning is performed line by line up to the lower left corner of the character rectangular image P801. Assuming that the number of vertical pixels of the image P801 exists up to an arbitrary number b of lines, a storage location for black pixel count numbers from Y1 to Yb is required.

  In scanning from the left side, counting of the number of black pixels ends when a white pixel appears, and thus scanning from the right side (−X direction) is also necessary. Therefore, a storage location for the black pixel count number from Y′1 to Y′b is also required. That is, in step S401, scanning is further performed from the upper right end of the character rectangular image P801 in the X direction (−X direction), and the number of black pixels that appear continuously from the first appearing black pixels is counted. If a white pixel appears, the scanning of that line ends. The number of pixels obtained at this time is stored in Y′1. Similarly, scanning is performed line by line up to the lower right corner of the character rectangular image.

  Subsequently, scanning in the Y direction from the origin (scanning from the upper side) is performed, and the number of black pixels appearing continuously from the first appearing black pixels is counted (step S402). If a white pixel appears, the scanning of that line ends. The number of pixels obtained at this time is stored in X1. Similarly, scanning is performed for each line up to the upper right end of the character rectangular image P801.

  Further, in step S402, scanning is performed in the Y direction (−Y direction) from the lower left end of the character rectangular image P801, and the number of black pixels continuously appearing from the first appearing black pixels is counted. If a white pixel appears, the scanning of that line ends. The scanning from the lower left corner may be a scanning from the lower right corner at the opposite end.

  In order to facilitate the description of the outer frame removal method, the 12 × 12 square image shown in FIG. 11B will be described as an example. One square in FIG. 11B represents one pixel, and the hatched square means a black pixel. In addition, the arbitrary number b of lines is 12 in this example. FIG. 11 (2) shows the distribution of black pixels constituting the square frame.

  As described above, when the black pixel count numbers from X1 to Xb, X′1 to X′b, Y1 to Yb, and Y′1 to Y′b are obtained, the results shown in FIGS. 12 (1) to 12 (4) are obtained. It becomes the table shown. Here, “t” in FIG. 12 is an index of X and Y, and is a number representing the number of scanning lines on the X axis and the Y axis.

  Next, the frame removal unit 9 calculates the average number of pixels for removing the outer frame from the black pixel count numbers in steps S401 and S402 (step S403). In the present embodiment, the average number of pixels is calculated by detecting a change point in the scanning result from each side (four sides).

  As can be seen from FIGS. 12 (1) to 12 (4), when the character rectangular image has a square shape, from the end (t = 1 and 12) to the center (t = 6 and 7) of the image. As it goes, the count number of continuous black pixels decreases. In addition, it can be seen that the difference in the black pixel counts of t = 1 and t = 2 and t = 12 and t = 11, or t = 2 and t = 3, and t = 11 and t = 10 is large. Such a line having a large difference in the black pixel count number can be considered as a change point from the square outer frame (black pixel) to the background area (white pixel).

  The calculation of the average number of pixels will be described with reference to FIG. The number of black pixel counts will be examined in order from t = 1. The difference between the black pixel count when t = 2 and the black pixel count when t = 1 is 11. The difference between the count numbers of black pixels at t = 2 and t = 3 is zero. Between t = 1 and t = 2 where the difference in the number of counts of black pixels is large, it is estimated that the change point from the outer frame of the square (black pixels) to the background area (white pixels). Also, the difference in the count number of black pixels is calculated from t = 12. Then, since the difference between t = 12 and t = 11 is 10, and the next difference between t = 11 and t = 10 is 0, the difference between t = 12 and t = 11 is large. Therefore, it is estimated that the point between t = 12 and t = 11 is a change point from the rectangular outer frame to the background region.

  Of the black pixel counts obtained in FIG. 12A, the average value of the black pixel counts sandwiched between the portions estimated (determined) as the change points is calculated. This value is the number of pixels removed from the outer frame. Here, it is calculated from the black pixel count from t = 2 to t = 11.

  FIGS. 13 (1) to (4) show the calculation results of the average values of the black pixel count numbers of FIGS. 12 (1) to (4). As shown in FIGS. 13 (1) to 13 (4), the average value of the black pixel count number is 2 in any of the scans from the four sides.

  When the average value of the number of black pixels is calculated, the frame removing unit 9 converts the black pixels for the average number of pixels into white pixels in order from the end (step S404). That is, the square frame is removed by inverting black pixels to white pixels by the calculated number of black pixels from each end of the four sides.

  FIG. 14 shows a state where the square frame is removed (the black pixels constituting the square frame are converted into white pixels) based on the calculation result of FIG. FIG. 14 (1) shows an enlarged view of the square frame shown in FIG. 11 (2).

  Referring to FIG. 14 (2), from the calculation result of FIG. 13 (1), the black pixels for two pixels are inverted into white pixels in the Y direction while scanning in the X direction from the upper left corner. At this time, if the black pixel has one pixel and the white pixel has one pixel, only one pixel is inverted to a white pixel. The results of performing the same processing in order based on the calculation results of FIGS. 13 (2) to 13 (4) are shown in FIGS. 14 (3) to 14 (5), respectively. When such a process is performed, as shown in FIG. 14 (5), all the black pixels constituting the square frame are removed. However, the order of conversion into white pixels is not limited to the order shown in FIG.

  Note that some characters may touch the square frame, and at this time, the black pixel count of the contact portion is expected to increase. In this case, it is preferable not to use a pixel having an extremely large number of black pixel counts compared to the number of black pixel counts in other scanning lines for the average value calculation.

  According to the present embodiment, the rectangular frame of the character rectangular image P801 can be removed by the above procedure. Details of the rectangular frame removal of the character rectangular image P801 shown in FIG. 7 are not described here, but can be removed by the above method.

  Even when the process for removing the square frame is performed, the original character rectangular image P801 is stored in the memory 15.

  Referring to FIG. 3 again, when the square frame is removed, matching unit 10 collates the remaining character pattern, that is, the character pattern from which the frame has been removed, with one-character dictionary 11 (step S207). The reason for performing such processing is as follows.

  If the character is a Chinese character, there is a Chinese character whose character shape will be determined as a square, such as “country, figure, stick,...”. For this reason, in the present embodiment, a one-character dictionary for registering characters in Kanji squares (such as Kuniga-mae) that will be determined as squares, such as “ball, old”, etc., as patterns. 11 is held in advance.

  If the rectangular frame of the character rectangular image P801 can be removed, the remaining character pattern is “place”. Actually, since it is a character image, the matching unit 10 recognizes that it is a character pattern in the form of “field”, and therefore collates this character pattern with the one-character dictionary 11. In the one-character dictionary 11, character patterns inside kanji having a square feature are registered in advance. In the present embodiment, for example, a plurality of character patterns (specific character patterns) as shown in FIG.

  If the character pattern after the frame removal is a character that does not exist in the one-character dictionary 11 (YES in step S208), the character recognition processing unit 12 applies the character pattern (the character pattern after the frame removal) to the character recognition processing. (Step S209). The text data obtained by this processing is temporarily stored in the memory 15. When this process ends, the process proceeds to step S211.

  In the example of the character rectangular image P801, the matching unit 10 collates the character pattern of the one-character dictionary 11 and the character pattern (“field” character pattern) P801 # after the frame removal one by one, and determines whether or not they match. Determine. Since the one-character dictionary 11 shown in FIG. 15 does not have a character pattern that matches the “place” character pattern P801 #, it is assumed that the character pattern P801 # is a character that does not exist in the one-character dictionary 11, and the process proceeds to step S209. Processing is performed.

  In step S209, since the character pattern P801 # from which the square frame has been removed is recognized, it is finally converted into text data “place” and temporarily stored in the memory 15.

  In step S208, if the character pattern after the frame removal matches the character pattern in the one-character dictionary 11 (NO in step S208), the process proceeds to step S210 described above, and the character recognition processing unit 12 sets the frame. The original character that has not been deleted is subjected to character recognition processing. The text data of the recognized character is temporarily stored in the memory 15.

  In step S211, it is determined whether or not the character recognition of all the character rectangular images in the line has been completed. If not completed (NO in step S211), the next character rectangular image is selected, and the process returns to step S204. If completed (YES in step S211), the process proceeds to step S212.

  In step S212, it is determined whether or not processing of all lines in the input image (character string image) has been completed (step S212). If not completed (NO in step S212), the next character string line is selected, and the process returns to step S203. If completed (NO in step S212), the process proceeds to step S213.

  In step S 213, text data after character recognition temporarily stored in the memory 15 is output to the output device 13. Thereby, the text data of the read character string image is displayed on the output device 13. In the case of the example of the character rectangular image P801 described above, the character rectangular image P801 is recognized as the character “place” and is output to the output device 13.

  Note that “place” may be stored in advance in association with the meaning of “place”, and “place” may be output as a recognition result. In addition, in recent mobile phones, etc., there is a function that can read a business card and register the company name and name, but by applying this method, in the registration item of “location” such as the phone book of the mobile phone, A mechanism for determining and registering a character string recognized after “no place” as an address or place of a store or the like is also conceivable. This mechanism is the same even when the kanji in the square frame is other than “ba”. In this embodiment, the “place” with a square is recognized. However, other characters with a square such as kanji and symbols can be recognized by the same process.

Through the above processing, the character string whose character has been recognized is displayed on the output device 13.
Therefore, even if various kinds of characters with squares (combined characters of arbitrary characters and square frames) are not registered in the character recognition dictionary 4 in advance, the characters surrounded by the square frames can be properly recognized.

  In the present embodiment, if the character after the removal of the square frame is a character that exists in the one-character dictionary 11, the square frame is not deleted, but actually, “ It is thought that there may be a case of “size” or “mouth”. Therefore, in order to recognize characters with higher accuracy, it is preferable to further execute the following processing.

  That is, when it is determined in S208 of FIG. 3 that the character is present in the one-character dictionary 11 (NO in step S208), the character with a square frame including the character existing in the one-character dictionary 11 is used as the character information. Information for identifying this is temporarily stored. Then, the character recognition of the entire page is performed once (in this case, the original character recognition without deleting the frame is performed for the character). Thereafter, a process of confirming the “accuracy” of the sentence including the character is performed in the form of post-processing (between steps S212 and S213).

  For example, in the case of “it is the first character of one sentence” or “it is a character in the middle of one sentence but there is a space before that character”, there is a high possibility that it is a composite character with a square frame. Therefore, if the character matches such a condition, the frame removal process may be performed again.

  Alternatively, a separate word dictionary may be held, and if the character string matches the word dictionary, the frame is not removed as it is, and if the character string does not match, the frame may be removed. Specifically, for example, if the sentence is “King is ...”, the word “King” is established if the frame is not removed, but “Tamao” is obtained if the frame is removed. Since the word like the latter does not make sense (it does not exist in the word dictionary), it can be determined that the original character is correct.

  In addition, when using characters with a square frame or a circle with a magazine, etc., it is considered that two or more such characters are often used. If there is no framed character, the frame may not be deleted.

  In addition, when the above process is added, the character after removing the frame may be first recognized and then the accuracy of the sentence may be determined.

  Alternatively, the character recognition result of both the original character from which the frame has not been removed and the character after the frame has been removed may be presented (for example, displayed) to allow the user to select which is correct.

  In the present embodiment, a pattern of characters (symbols) inside a kanji that would be determined to be a quadrangle is registered in the one-character dictionary 11, but it is determined to be a quadrangle. The pattern of the kanji (such as “country”) itself that will be stored may be registered in advance. In this case, the process of step S206 (removal of the square frame) is performed when it is determined in step S208 that the character exists in the one-character dictionary 11 (NO in step S208).

  Or conversely, a method may be used in which a kanji character to be determined as a character with a square (composite character of a square frame and a character) is registered in a one-character dictionary. However, there are about 6400 JIS level 1 and level 2 characters for Japanese Kanji characters, and there are almost no Kanji characters that cannot be used to create square characters. It is considered more practical to register.

[Embodiment 2]
Next, a method for recognizing characters surrounded by a round frame will be described.

  The configuration and basic operation of the information processing apparatus in the present embodiment are the same as those in the first embodiment. Only the parts different from the first embodiment will be described below.

  In the present embodiment, the frame removing unit 9 calculates the number of black pixels continuous from the ends of two opposing sides by scanning the rectangular image in the vertical direction or horizontal direction. Then, the outer frame is removed by inverting black pixels to white pixels by the calculated number of black pixels from the ends of the two opposite sides.

  FIG. 16 is a flowchart showing character recognition processing executed by the information processing apparatus 1 according to Embodiment 2 of the present invention. The process shown in the flowchart of FIG. 16 is also stored in advance in the HDD 212 as a program, and the function of the character recognition process is realized by the CPU 210 reading and executing this program. In addition, the same step number is attached | subjected about the process similar to the process shown in FIG. Therefore, description thereof will not be repeated.

  Compared with Embodiment 1, in this embodiment, instead of Steps S204, S205, and S206, the processes of Steps S1204, S1205, and S1206 are executed. In the present embodiment, the processes in steps S207 and S208, which are processes related to the one-character dictionary 11, are deleted. Accordingly, in the present embodiment, the matching unit 10 and the one-character dictionary 11 are not essential.

  In step S1204, the character shape determination unit 8 selects one extracted character rectangular image, and determines whether the shape of the image is a circle. That is, it is determined whether or not the character shape is a circle from the distribution of black and white pixels in the character rectangular image. The determination here may also be performed by a known method such as Hough transform.

  Note that the character segmentation process (step S203) in the present embodiment may be performed by a known method, but the outermost black pixels on the upper, lower, left, and right sides of the black pixels of each character image are set to the maximum coordinates. Conditioned to cut out.

  In step S1205, it is determined whether or not the character shape is determined to be a circle. If it is determined to be a circle (YES in step S1205), the process proceeds to step S1206. If it is determined not to be a circle (NO in step S1205), the process proceeds to step S210 described above.

  Here, as shown in FIG. 17, a character rectangular image P1601 indicating “when a circle (◯) is attached” in the character string row P704 (FIG. 6) is selected, and it is determined that the image P1601 has a round shape. This will be described below.

  In step S1206, the frame removal unit 9 executes a process for removing a round frame (hereinafter also referred to as “round frame removal process”).

  FIG. 18 is a flowchart showing the round frame removal process. FIG. 19 is a diagram for explaining the round frame removal process. FIGS. 19A and 19B are enlarged views of the character rectangular image P1601. The dotted lines in FIGS. 19 (1) and (2) represent the cut out rectangular area. Similarly to the first embodiment, the upper left corner of the character rectangular image P1601 is the origin, the horizontal direction is the X direction, and the vertical direction is the Y direction.

  First, the outline will be described. First, scanning is performed in the X direction from the origin of the character rectangular image P1601, and the number of black pixels is counted. From the start of scanning, the number of continuous black pixels from the first appearance of black pixels to the appearance of white pixels is counted. In the case of a square, counting was performed from the first appearance of a black pixel, but in the case of a circle, there are many background images (white pixels) from the origin to the outer frame of the circle. Adopt one that has

  Here, in order to facilitate the description of the method of removing the round frame, description will be made using the 12 × 12 square image shown in FIG. One square in FIG. 20 represents one pixel, and a hatched square means a black pixel. FIG. 20 shows the distribution of black pixels constituting the round frame.

  Referring to FIGS. 18 and 20, first, scanning is performed line by line in the X direction from the origin, and the number of black pixels that appear continuously is counted (step S1511). Here, as shown in FIG. 19B, scanning from the origin to the X direction is performed up to “K”. “K” is a value half the number of pixels on the X-direction axis (X-axis). At this time, scanning is similarly performed from the upper right end toward the position “K” in the X direction (actually −X direction). In addition, the count numbers of the black pixels that appear continuously are stored in “Y1 to Yb” and “Y′1 to Y′b”, respectively. “B” is an arbitrary integer, and here represents the number of lines to be scanned.

  The results of scanning the rectangular image of FIG. 20 and counting the number of black pixels are shown in the tables of FIGS. 21 (1) and (2). The count result of FIG. 21 (1) corresponds to the left half region R1 with reference to the center position (K) in the X direction in FIG. 20, and the count result of FIG. , Corresponding to the right half region R2 with reference to the center position (K) in the X direction.

  Here, “b” is 12 because it is the number of lines to be scanned, and “t” is an index of Y. Y1 is the black pixel count when t = 1. In the case of t = 1 shown in FIG. 21A, scanning is performed from the origin, but the first pixel FPX shown in FIG. 20 is a cluster of only one pixel (the next pixel is a white pixel). Therefore, the first pixel FPX is not counted, and further scanning is performed until the next black pixel appears. As shown in FIGS. 21 (1) and (2), for each line, the position where counting is started (the number from the left end or the right end) is stored in association with the number of black pixels.

  Next, the frame removal unit 9 obtains an average value from the black pixel count numbers in FIGS. 21 (1) and 21 (2) (step S1512). For each of FIGS. 21A and 21B, when the total number of black pixel counts is divided by the number of t (here, 12), “2” is calculated as an average value.

  When the average value is calculated, scanning is performed line by line in the Y direction from the upper left end and the upper right end of the image, and the pixels for two pixels in the X direction are inverted to white pixels (step S1513). This process will be specifically described with reference to FIG.

FIG. 22 (1) shows an enlarged view of the round frame shown in FIG.
Referring to FIG. 22 (2), first, when scanning from the upper left end (origin), the first pixel is a black pixel in the first line, but as shown in FIG. Since the start position is the fourth pixel, two pixels from the fourth black pixel are inverted to white pixels. That is, since the first pixel is excluded from the black pixel count in step S1511, it is ignored (skip), and two pixels from the next black pixel appearing are inverted to white pixels.

  Here, if the number of pixels for the black pixel removal is advanced, if it is larger than “K” (here, 6), only the processing up to “K” is performed.

  Thereafter, scanning is similarly performed to the lower left corner, and two pixels are inverted to white pixels. As a result, an image as shown in FIG. 22 (2) is obtained.

  Next, scanning is performed in the X direction (actually in the negative X direction) from the upper right end of the image in FIG. 22B, and two pixels are inverted to white pixels from the count start position. As a result, an image as shown in FIG. 22 (3) can be obtained.

  As shown in FIG. 22 (3), when the process of inverting the number of average black pixels is performed, some black pixels may remain. However, if a known method for removing isolated points in image processing is used, the remaining points can be erased.

By the method as described above, the outer frame (round frame) at the time of “rounded” can be removed.
Referring to FIG. 16 again, when the round frame is removed, the process proceeds to step S209 described above, and the character rectangular image (character pattern “hour”) after the round frame is removed is subjected to the character recognition process. Thereby, text data “time” can be obtained.

  As described above, even if it is a circled character (a composite character of an arbitrary character and a circle frame), even if various types of circled characters themselves are not registered in the character recognition dictionary 4 in advance, The enclosed characters can be recognized properly.

  As in the first embodiment, if “hour” is recognized, the numbers on the same line recognized next to the characters may be recognized as time.

  Moreover, you may combine the process of Embodiment 1 and Embodiment 2. FIG. That is, in one information processing apparatus, both the removal of the square frame and the removal of the round frame may be performed.

  In the first and second embodiments, the outer frame is described as a square frame or a round frame. However, the outer frame is not limited to these, and the outer frame is an ellipse, a polygon, a rectangle, or the like. However, it is possible to recognize characters in the outer frame by the same processing.

  In the first embodiment, the description has been made only in the case where a single character is included in the outer frame. However, by preparing a pattern for a plurality of characters in a one-character dictionary, a plurality of characters surrounded by the outer frame are provided. Characters can also be recognized.

  The character recognition method performed by the information processing apparatus of the present invention can also be provided as a program. Such a program can be recorded on an optical medium such as a CD-ROM (Compact Disc-ROM) or a computer-readable recording medium such as a memory card and provided as a program product. A program can also be provided by downloading via a network.

  The program according to the present invention is a program module that is provided as a part of a computer operating system (OS) and calls necessary modules in a predetermined arrangement at a predetermined timing to execute processing. Also good. In that case, the program itself does not include the module, and the process is executed in cooperation with the OS. A program that does not include such a module can also be included in the program according to the present invention.

  The program according to the present invention may be provided by being incorporated in a part of another program. Even in this case, the program itself does not include the module included in the other program, and the process is executed in cooperation with the other program. Such a program incorporated in another program can also be included in the program according to the present invention.

  The provided program product is installed in a program storage unit such as a hard disk and executed. Note that the program product includes the program itself and a storage medium in which the program is stored.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

It is a functional block diagram showing the function structure of the information processing apparatus in each embodiment of this invention. It is a hardware block diagram of the information processing apparatus in each embodiment of the present invention. It is a flowchart which shows the character recognition process which the information processing apparatus which concerns on Embodiment 1 of this invention performs. It is a figure which shows an example of a character string image (input image). It is a figure which shows an example of the character string image after binarization. It is a figure which shows the example from which the character string line was cut out from the image after binarization. It is a figure which shows an example of the character rectangular image cut out from the character string line. It is a flowchart which shows the flow of a square determination process. It is a figure for demonstrating the square determination method. It is a flowchart which shows a square frame removal process. It is a figure for demonstrating a square frame removal process. (1) to (4) are tables showing black pixel count numbers obtained by scanning from four sides. 12A and 12B are diagrams showing the calculation results of the average value of the black pixel count numbers. It is a figure which shows a mode that a square frame is removed based on the calculation result of FIG. It is a figure which shows the concept of the matching process of the specific character pattern registered into the 1 character dictionary, and the character pattern after a frame removal. It is a flowchart which shows the character recognition process which the information processing apparatus which concerns on Embodiment 2 of this invention performs. It is a figure which shows an example of a round-shaped character rectangular image. It is a flowchart which shows a round frame removal process. It is a figure for demonstrating a round frame removal process. It is a figure which shows the example of distribution of the black pixel which comprises a round frame. (1) and (2) are tables showing black pixel count numbers and black pixel number average values obtained by scanning from opposite sides, respectively. It is a figure which shows a mode that a round frame is removed based on the calculation result of FIG.

Explanation of symbols

  1 information processing device, 2 input device, 3 control unit, 4 character recognition dictionary, 5 binarization processing unit, 6 character string row cutout unit, 7 character cutout unit, 8 character shape determination unit, 9 frame removal unit, 10 matching 11, 1 character dictionary, 12 character recognition processing unit, 13 output device, 14 input / output control unit, 15 memory, 210 CPU, 211 operation unit, 212 HDD, 214 memory unit, 215 CD-ROM drive, 216 input / output I / F, 217 Communication interface unit, 218 Internal bus, 221 Image reading device, 222 Display device.

Claims (12)

An information processing apparatus for recognizing characters surrounded by an outer frame from a character string image,
Character cutout means for cutting out a rectangular image corresponding to one character from the character string image;
Shape determining means for determining whether or not the character shape in the rectangular image matches a predetermined type of outer frame shape based on the distribution of monochrome pixels in the cut out rectangular image;
When it is determined by the shape determining means that the images match , the width of the outer frame to be removed is determined based on the distribution of black pixels within the range corresponding to the outer frame shape determined to match, and the determined width based on the frame removing means for removing an outer frame of the character in said rectangular image,
An information processing apparatus comprising: recognition means for performing character recognition on the character pattern of the image after processing by the frame removal means. Determines that the frame removing means, said a plurality of scanning in a first direction parallel to one side of the rectangular image, the number of consecutive black pixels from the end portion side of the rectangular image is calculated for each scan, the match In accordance with the outer frame shape calculated, an average value of the number of black pixels calculated in all or part of the plurality of scans is calculated, and black pixels are inverted into white pixels by the number of black pixels of the calculated average value. The information processing apparatus according to claim 1, wherein the outer frame is removed. The outer frame shape of the predetermined type includes a round shape,
The frame removing means includes
When the shape determining means determines that the character shape in the rectangular image is the round shape, the scanning of the rectangular image in the first direction or in the second direction orthogonal to the first direction, Means for calculating the number of continuous black pixels from the ends of two opposing sides;
And means for removing an outer frame by inverting black pixels to white pixels by the number of black pixels of the average value calculated from the ends of the two opposing sides. Information processing device. The outer frame shape of the predetermined type includes a square shape,
Specific character storage means for storing in advance a character pattern inside a character whose outside is a square frame as a specific character pattern;
Matching means for collating the character pattern of the image processed by the frame removing means with the specific character pattern when the shape determining means determines that the character shape in the rectangular image is the quadrangular shape And
The recognition means performs character recognition of the original character pattern with the outer frame remaining cut out by the character cutout means when the matching means determines that it matches the specific character pattern. The information processing apparatus according to any one of 2 and 3.   The information processing apparatus according to claim 4, wherein the recognizing unit performs character recognition of the character pattern processed by the frame removing unit when the matching unit determines that the specific character pattern does not match. A storage means,
The storage unit holds both the image after the processing by the frame removing unit and the rectangular image before the outer frame is removed when it is determined by the shape determining unit that they match. Or the information processing apparatus according to 5. The outer frame shape of the predetermined type includes a square shape,
Specific character storage means for preliminarily storing a character pattern whose outside is a square frame as a specific character pattern;
A matching means for matching the character pattern of the rectangular image with the specific character pattern when the shape determining means determines that the character shape in the rectangular image is the square shape;
4. The information according to claim 2, wherein the frame removing unit removes an outer frame of a character in the rectangular image only when the matching unit determines that the character pattern does not match the specific character pattern. 5. Processing equipment. The frame removing means includes
When the shape determining means determines that the character shape in the rectangular image is the quadrangular shape, the scanning of the rectangular image in the first direction and the second direction orthogonal to the first direction, Means for calculating the number of continuous black pixels from the ends of the four sides of the rectangular image;
A means for removing an outer frame by inverting black pixels to white pixels by the calculated number of black pixels corresponding to the average value from each end of four sides. The information processing apparatus described in 1.   The information processing apparatus according to any one of claims 1 to 8, wherein the character surrounded by the outer frame includes any one of a kanji character, an English character, a hiragana character, a number, and a katakana character in the frame. A method for recognizing characters enclosed in an outer frame from a character string image,
Cutting out a rectangular image corresponding to one character from the character string image;
Determining whether a character shape in the rectangular image matches a predetermined type of outer frame shape based on a distribution of black and white pixels in the cut out rectangular image;
When it is determined that the character shape in the rectangular image matches the outer frame shape of the predetermined type, the outer shape to be removed based on the distribution of black pixels within the range corresponding to the outer frame shape determined to match Determining a width of the frame, and removing an outer frame of characters in the rectangular image based on the determined width ;
Performing character recognition on the character pattern of the image after removing the outer frame. A program for recognizing characters enclosed in an outer frame from a character string image,
Cutting out a rectangular image corresponding to one character from the character string image;
Determining whether a character shape in the rectangular image matches a predetermined type of outer frame shape based on a distribution of black and white pixels in the cut out rectangular image;
When it is determined that the character shape in the rectangular image matches the outer frame shape of the predetermined type, the outer shape to be removed based on the distribution of black pixels within the range corresponding to the outer frame shape determined to match Determining a width of the frame, and removing an outer frame of characters in the rectangular image based on the determined width ;
A character recognition program for causing a computer to execute character recognition for a character pattern of an image after removing an outer frame.   The computer-readable recording medium which recorded the character recognition program of Claim 11.

Images