WO2013108347A1 - Character recognition device, classifying device provided with same, character recognition method and control program - Google Patents

Character recognition device, classifying device provided with same, character recognition method and control program Download PDF

Info

Publication number
WO2013108347A1
WO2013108347A1 PCT/JP2012/008210 JP2012008210W WO2013108347A1 WO 2013108347 A1 WO2013108347 A1 WO 2013108347A1 JP 2012008210 W JP2012008210 W JP 2012008210W WO 2013108347 A1 WO2013108347 A1 WO 2013108347A1
Authority
WO
WIPO (PCT)
Prior art keywords
character
recognition
word
unit
registered
Prior art date
Application number
PCT/JP2012/008210
Other languages
French (fr)
Japanese (ja)
Inventor
大輔 西脇
Original Assignee
日本電気株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電気株式会社 filed Critical 日本電気株式会社
Publication of WO2013108347A1 publication Critical patent/WO2013108347A1/en

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C3/00—Sorting according to destination
    • B07C3/10—Apparatus characterised by the means used for detection ofthe destination
    • B07C3/14—Apparatus characterised by the means used for detection ofthe destination using light-responsive detecting means
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/768—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using context analysis, e.g. recognition aided by known co-occurring patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/14—Image acquisition
    • G06V30/148—Segmentation of character regions
    • G06V30/153—Segmentation of character regions using recognition of characters or words
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition

Definitions

  • the present invention relates to a character recognition device that reads characters printed or written on a paper surface, a sorting device including the character recognition device, a character recognition method, and a control program.
  • mail items are scanned by scanning images of mail using an image scanner, etc., and reading destination information such as address and address from the scanned image using OCR (Optical Character Recognition) technology. This is done by sorting mail items based on destination information.
  • OCR Optical Character Recognition
  • Patent Document 1 discloses a technique related to classification of mail items.
  • a technique for reading destination information from a scanned image is disclosed in Non-Patent Document 1, for example.
  • reading of destination information based on a scanned image can be performed by extracting features of the scanned image, and this feature extraction method is disclosed in Non-Patent Document 2, for example.
  • the scanned image is decomposed into character images for each character, input to the engine for recognizing each decomposed character image to acquire a character code as a recognition result, and destination information by combining the acquired character codes As the output method.
  • the second is a method in which the scanned image is directly input to the recognition engine without being decomposed into a character image, and the recognition result of the scanned image itself is output as destination information.
  • the scanned image is recognized as it is without being decomposed into a character image, it is only necessary to determine which of the destination information is registered in advance, and the probability that erroneous destination information is output is low. For example, when reading the address of Tokyo 23 wards, it is only necessary to select one out of 23 candidates. However, when a scanned image is recognized as it is, the recognition processing load increases, and if the candidate cannot be narrowed down to one, recognition is impossible.
  • an object of the present invention is to provide a character recognition device, a sorting device including the character recognition device, a character recognition method, and a control program that can recognize words with high accuracy and low probability of being unrecognizable. There is.
  • a character recognition device includes a word table in which a plurality of words are registered in association with a predetermined character string, and image information of a display body on which a recognition target is displayed.
  • Recognizing means for extracting a string and outputting a word associated with the extracted character string as a recognition target recognition result.
  • a sorting apparatus reads a piece of image information from a conveying device that conveys a shipment on which a recognition target is displayed, and a shipment being conveyed, and outputs a recognition result of the recognition target.
  • the above character recognition device and a sorting device that sorts the shipments that have been conveyed based on the output recognition result.
  • a character recognition method is a character recognition using a word table in which a plurality of words are registered in association with a predetermined character string, and a recognition target is displayed.
  • the image information of the display body is read, the read image information is divided into unit images, the divided unit images are analyzed and arranged in the order of division, and output as analysis results, and character strings are extracted from the word table based on the analysis results.
  • the extracted word string is output as a recognition target recognition result.
  • a control program is a control program executed by a computer of a character recognition device having a word table in which a plurality of words are registered in association with a predetermined character string.
  • This control program has the function of reading the image information of the display object on which the recognition target is displayed, the function of dividing the read image information into unit images, and analyzing the divided unit images and arranging them in the order of division and outputting them as analysis results. And a function of extracting a character string from the word table based on the analysis result and outputting a word associated with the extracted character string as a recognition target recognition result.
  • the character recognition device including the character recognition device, the character recognition method, and the control program can recognize words with high accuracy with a low probability of being unrecognizable.
  • FIG. 1A shows a configuration diagram of the sorting apparatus according to the present embodiment
  • FIG. 1B shows a block configuration diagram of the character recognition device.
  • the sorting apparatus 10 includes a transport device 20, a character recognition device 30, and a sorting device 40.
  • the transport device 20 transports the shipment 50 (display body) on which the destination information that is the recognition target is displayed.
  • the character recognition device 30 captures image information of the shipment 50 being conveyed by scanning or the like, recognizes destination information from the captured image information, and outputs the recognized destination information to the sorting device 40.
  • the sorting device 40 sorts the shipped items 50 that have been transported based on the destination information output from the character recognition device 30.
  • the character recognition device 30 includes a word table 31, a reading unit 32, a sorting unit 33, an analysis unit 34, and a recognition unit 35.
  • a plurality of words are registered in association with a predetermined character string.
  • all destination information necessary to classify the shipment 50 is registered as words, and a character string for extracting the destination information is associated with each destination information. .
  • the reading means 32 reads the image information of the shipment 50 on which the recognition target is displayed.
  • a scanner can be applied as the reading unit 32.
  • the sorting unit 33 sorts the image information read by the reading unit 32 from the shipment 50 into character images (unit images) for each character, and adds the order (classification order) in which the image information was formed to analyze the image information. 34.
  • the analyzing unit 34 calculates the feature amount of the character image output from the sorting unit 33 and specifies each character having the calculated feature amount.
  • the analysis unit 34 further generates a character string by arranging the specified characters in the order of division, and outputs the character string to the recognition unit 35 as an analysis result.
  • the recognition unit 35 extracts a character string that matches the analysis result output from the analysis unit 34 from the word table 31, and outputs destination information (word) associated with the extracted character string as a recognition result.
  • the character recognition device 30 and the sorting device 10 configured as described above classify the image information of the shipment 50 into predetermined character images (unit images), and analyze the character images respectively.
  • the analysis process is simple and a character can be specified with high accuracy.
  • the character recognition device 30 and the sorting device 10 include a word table 31 in which a plurality of pieces of destination information (words) are registered in association with character strings, and characters (analysis) output in the order of sorting. A character string that matches (result) is extracted from the word table 31, and destination information (word) associated with the extracted character string is output as a recognition result.
  • the word recognition process can be replaced with a process of extracting one of the finite character string candidates, which can reduce the possibility of being unrecognizable and output incorrect destination information (word) as a recognition result. Can be reduced.
  • the character recognition device 30 and the sorting device 10 have a low probability of being unrecognizable and can recognize the recognition target with high accuracy, and appropriately select the shipment 50 on which the recognition target is displayed. Can be sorted.
  • the character recognition device 30 extracts the character string closest to the analysis result.
  • the edit distance can be used to extract the closest character string. The extraction of the character string using the edit distance will be described in detail in the second embodiment.
  • FIG. 2 shows a block diagram of the character recognition device according to this embodiment.
  • the data flow is indicated by a solid line
  • the control signal flow is indicated by a dotted line.
  • the character recognition device 100 includes an input module 110, a character segmentation module 120, a feature extraction module 130, a character classification module 140, a classification table 150, a conversion module 160, a conversion table 170, and an output.
  • a module 180 and a control module 190 are provided.
  • the input module 110 obtains a word image from the magazine, for example, when the user sets a magazine on which the recognition target is printed on the image reading stand and presses the start button, and the obtained word image is converted into the word image data. Is output to the character segmentation module 120.
  • a scanner or the like can be applied as the input module 110.
  • the input module 110 acquires a word image, for example, by scanning a specific area of a magazine, and outputs the word image data “Abc” to the character segmentation module 120.
  • the character cutout module 120 decomposes the word image data output from the input module 110 into character image data for each character and outputs the character image data to the feature extraction module 130.
  • the character segmentation module 120 for example, decomposes the word image data “Abc” output from the input module 110 into character image data for each character, and three character images “A”, “b”, and “c”. Output data.
  • the feature extraction module 130 performs image processing on the character image data output from the character cutout module 120, extracts the feature amount of each character image data, determines character elements, and outputs them to the character classification module 140.
  • the feature extraction module 130 can use a feature amount extraction method generally used in image processing in an optical character recognition device (OCR). In the feature amount extraction method, for example, the direction information of the edge of the character for each local region disclosed in Non-Patent Document 2 shown in the background art can be used.
  • OCR optical character recognition device
  • the feature extraction module 130 processes, for example, three character image data “A”, “b”, and “c”, respectively, and determines “character element A”, “character element b”, and “character element c”. And output to the character classification module 140.
  • the character classification module 140 refers to the classification table 150 to identify the character group to which the character element output from the feature extraction module 130 belongs. Further, the character classification module 140 generates an ID string by arranging the group IDs of the identified character groups in the order of output, and outputs the generated ID string to the conversion module 160.
  • classification table 150 all the character elements constituting the word to be recognized are classified and registered.
  • classification table 150 character elements having similar feature amounts are grouped, and a group ID is assigned to each group. Note that even if feature quantities are close to each other, if it is necessary to distinguish them, it is desirable to make them separate groups.
  • the classification table 150 corresponds to a character table in the claims.
  • FIGS. 3A and 3B An example of the classification table 150 is shown in FIGS. 3A and 3B.
  • FIG. 3A shows an English character classification table 150
  • FIG. 3B shows a Thai character classification table 150.
  • 3A and 3B English characters and Thai characters having similar feature amounts are grouped, and a group ID is assigned to each character group. For example, in FIG. 3A, three character elements “c”, “C”, and “G” having similar feature amounts are grouped and given a group ID “3”.
  • the classification table 150 shown in FIG. 3A and FIG. 3B all the characters and numeric character elements constituting English and Thai are registered, but they are used for the words registered in the conversion table 170 described later. Non-letter and numeric character elements may not be registered in the classification table 150.
  • the character classification module 140 identifies, for example, the character group to which the “character element A” output from the feature extraction module 130 belongs by referring to the classification table 150, and acquires the group ID “25” of the identified character group. To do. Further, the character classification module 140 acquires the group ID “2” of the character group to which “character element b” belongs, and the group ID “3” of the character group to which “character element c” belongs. Then, the character classification module 140 generates the ID string “25-2-3” by arranging the acquired group IDs in the order of output.
  • the conversion module 160 refers to the conversion table 170 to determine a word corresponding to the ID string output from the character classification module 140, and outputs the determined word to the output module 180.
  • the conversion table 170 a plurality of words are registered in association with the ID string.
  • the conversion table 170 corresponds to the word table in the claims.
  • An example of the conversion table 170 is shown in FIG. In FIG. 4, for example, an ID string “25-2-3 36-14-1-27” is associated with the word “Abc Road”.
  • the conversion module 160 extracts an ID string that matches the ID string “25-2-3 36-14-1-27” output from the character classification module 140 from the conversion table 170, and associates the ID string with the extracted ID string.
  • the attached word “Abc Load” is output to the output module 180.
  • the conversion module 160 extracts the closest ID string from the conversion table 170 when an ID string that matches the ID string output from the character classification module 140 is not registered in the conversion table 170. For example, in the character recognition of “Abc Road”, if the character classification module 140 misrecognizes the second “o” of “Load” as “e”, the character classification module 140 receives the ID string “25-2- ID string “25-2-3 36-5-1-27” is output instead of “3 36-14-27”. In this case, since the ID column “25-2-3 36-5-1-27” is not registered in the conversion table 170, the conversion module 160 uses the ID column “25-2-3 36-5-1-27”. ”Cannot be extracted.
  • the conversion module 160 uses the distance scale to obtain the ID closest to the ID string output from the character classification module 140.
  • a column is extracted from the conversion table 170. That is, when the ID string is considered as a character string, the closeness between the character strings can be defined as a cost for converting from one character string to the other character string. Costs are defined for insertion, deletion, and replacement of characters constituting the ID string, and the total cost required for conversion is considered as the distance between the ID strings. The ID string having the smallest distance between ID strings is taken as the corresponding ID string.
  • edit distance Longstein distance, edit distance
  • the ID string “25-2-3 36-14-1” described above is used.
  • the distance between ID columns of “ ⁇ 27” and the ID column “25-2-3 36-5-1-27” is 1 because one character is replaced. Note that if a predetermined threshold value is set and an ID string having an ID string distance smaller than the threshold value is not found, no recognition result can be set (rejected).
  • the output module 180 outputs the word “Abc Load” output from the conversion module 160 as the recognition result of the recognition target printed on the magazine.
  • the control module 190 controls operations of the input module 110, the character segmentation module 120, the feature extraction module 130, the classification module 104, the conversion module 160, and the output module 180, respectively.
  • Each of the modules described above has an input unit, a processing unit, a storage unit, and an output unit, and generates digital image data by scanning a magazine or a general computer that performs the above operation by processing digital data. It can be realized by a scanner or the like.
  • FIG. 5 shows an operation flow of the character recognition apparatus 100 according to this embodiment.
  • the input module 110 scans a specific area of the magazine and acquires a word image. .
  • the input module 110 outputs the acquired word image as word image data to the character segmentation module 120 (step S101).
  • the character cutout module 120 decomposes the word image data output from the input module 110 into character image data for each character and outputs the character image data to the feature extraction module 130.
  • Step S102 As a method of dividing character image data for each character, for example, for word image data, a histogram is created by projecting pixel values along the direction of the character string, and one character is defined between the valleys of the histogram. There is a way. Alternatively, it is possible to apply a labeling method in which an image for each character is created by examining connected regions of pixels and integrating adjacent connected regions.
  • the feature extraction module 130 performs image processing on the character image data output from the character cutout module 120, and outputs the character elements determined by extracting the feature amount of each character image data to the character classification module 140 (step S103). ).
  • the character classification module 140 refers to the classification table 150, specifies each character group to which the character element output from the feature extraction module 130 belongs, and acquires the group ID of the specified character group.
  • the character classification module 140 arranges the acquired group IDs in the output order of the feature values, and outputs them to the conversion module 160 as an ID string (step S104).
  • the conversion module 160 refers to the conversion table 170, determines a word associated with the ID string output from the character classification module 140, and outputs the determined word to the output module 180 (step S105).
  • the conversion module 160 uses the distance scale to select the ID string closest to the ID string output from the character classification module 140 as the conversion table 170. And the word associated with the extracted ID string is output.
  • the output module 180 outputs the word output from the conversion module 160 as a recognition result for the word image described in the specific area of the magazine (step S106).
  • the character recognition device 100 decomposes the word image data into character image data for each character, and specifies a character group belonging to each character image data.
  • the recognition process is simple and a character group to which the character group belongs can be specified with high accuracy.
  • the character recognition device 100 includes a conversion table 170 in which a plurality of words are registered in association with an ID string, and an ID string configured by a group ID of the specified character group is converted. It is determined which of the words registered in the table 170 corresponds to.
  • the word recognition process can be replaced with a selection process for selecting one of the finite candidates, so that the possibility of being unable to recognize can be reduced and the possibility of outputting an incorrect word recognition result can be reduced.
  • the selection process for selecting one of the finite candidates has a smaller calculation load than the word recognition process, so that the processing speed can be improved.
  • the character recognition device 100 can recognize words with high accuracy and a low probability of being unrecognizable.
  • the classification table 150 only letters and numbers constituting the words registered in the conversion table 170 may be registered. In this case, the arithmetic processing of the character classification module 140 can be minimized, and the word recognition processing speed can be further improved.
  • similar character elements that are less necessary to be distinguished are grouped.
  • word recognition can be performed without being affected by the similar characters.
  • the “C” and “c” in the classification table 150 belong to the same group, so that the conversion table 170 is referred to, and “ The correct word recognition result of “Abc” is output.
  • both “AbC” and “Abc” are registered in the conversion table 170, that is, when it is necessary to distinguish between both, “C” and “c” are different from each other in the classification table 150. Register with the group.
  • a plurality of character elements are registered in one character group, but the present invention is not limited to this.
  • a representative character element of a character group, an average of feature amounts of character elements belonging to the character group, and the like can be registered.
  • principal axes eigenvectors
  • the character classification module 140 sequentially compares the feature amount of the character image data output from the feature extraction module 130 with the feature amount of the character element of the character group registered in the classification table 150, and the closest feature amount. Identify character groups that have
  • FIG. 6 shows a configuration diagram of the shipment sorting machine according to the present embodiment.
  • the flow of data and control signals is indicated by a solid line
  • the flow of shipments is indicated by a broken line.
  • the shipment sorting machine 200 according to the present embodiment includes a supply unit 210, a transport unit 220, a scanner unit 230, an address recognition unit 240, a sorting unit 250, and a control unit 260.
  • the supply unit 210 supplies the shipments on which the destination addresses are displayed one by one to the transport unit 220.
  • the transport unit 220 transports the shipment supplied from the supply unit 210 to the sorting unit 250.
  • the scanner unit 230 captures an image of an area where the destination address of the shipment being transported by the transport unit 220 is displayed, and outputs the acquired captured image to the address recognition unit 240.
  • the address recognition unit 240 includes a classification table and a conversion table (not shown), refers to the classification table and the conversion table, recognizes the destination address from the captured image output from the scanner unit 230, and sorts the recognized destination address into the sorting unit 250. Output to.
  • the address recognition unit 240 determines an address area from the captured image output from the scanner unit 230, acquires a captured image of the determined address area as word image data, and configures the acquired word image data as an address. It has the function of decomposing into address words and the function of the character recognition device 100 described in the second embodiment.
  • the method currently disclosed by the nonpatent literature 1 shown to background art is applicable to the method of acquiring an address word from the captured image of a shipment thing, for example.
  • the address recognition unit 240 acquires an address word from the captured image of the shipment output from the scanner unit 230 and divides it into word image data.
  • the address recognition unit 240 calculates the feature amount of each divided word image data, identifies the character group to which the character element determined based on the calculated feature amount belongs, generates an ID string, and corresponds to the generated ID string.
  • the destination address to be selected is selected from the conversion table and output.
  • the sorting unit 250 includes a sorting table in which a destination address and a sorting box number are registered in association with each other.
  • the sorting unit 250 extracts the sorting box number associated with the destination address output from the address recognition unit 240 from the sorting table, and identifies the shipment that has been transported by the transporting unit 220 by the extracted sorting box number. Sort into separate boxes.
  • An example of the sorting table is shown in FIG. In the sorting table shown in FIG. 7, an address, an address, and a sorting box number are associated and registered.
  • the control unit 260 controls the operations of the supply unit 210, the transport unit 220, the scanner unit 230, the address recognition unit 240, and the sorting unit 250, respectively.
  • the supply unit 210, the conveyance unit 220, and the sorting unit 250 constituting the shipment sorter 200 are mechanical mechanisms for handling the shipment, and the scanner unit 230 is a photoelectric conversion mechanism that captures an image of the shipment. It is. These mechanisms can be realized by a general apparatus such as a mail sorting machine disclosed in Patent Document 1 shown in the background art.
  • the supply unit 210 supplies the shipments one by one to the transport unit 220 so that the area where the destination address of the shipment is displayed faces the scanner unit 230 side.
  • the scanner unit 230 acquires a captured image of an area where the destination address of the shipment is displayed while the shipment is being conveyed by the conveyance unit 220 to the sorting unit 250, and outputs the captured image to the address recognition unit 240.
  • the address recognition unit 240 decomposes the captured image into character image data, and calculates a feature amount for each character image data.
  • the address recognition unit 240 groups all characters (39 characters) used in 23 ward names such as “Ota Ward”, “Minato Ward”, “Chuo Ward”, and similar characters related to these 39 characters.
  • the character group to which the character element determined based on the calculated feature amount belongs is specified with reference to the classification table.
  • the address recognition unit 240 further generates an ID string by arranging the group IDs of the identified character groups in the display order, and refers to the conversion table in which 23 ward names and the ID string are associated and registered.
  • the destination address associated with the column is output to the sorting unit 250. Since “Ota Ward” is registered in the conversion table in association with the ID column, the destination address of the shipment that is misprinted as “Ota Ward” is correctly recognized as “Ota Ward”.
  • the sorting unit 250 sorts the shipped items that have been conveyed into a sorting box corresponding to the destination address recognized by the address recognition unit 240 by referring to the sorting table.
  • a shipment that is mistyped as “Ota Ward” is appropriately sorted into a sorting box corresponding to “Ota Ward”.
  • the shipment sorting apparatus 200 obtains character image data for each character from the captured image of the shipment, calculates the feature amount of each character image data, and determines the character element. As a result, an ID string is generated.
  • the recognition process is performed for each character image data for each character, the processing load is small and the ID string can be generated with high accuracy.
  • the shipment sorting machine 200 extracts the destination address corresponding to the generated ID string from the conversion table and outputs it. In this case, it is possible to reduce the possibility that output is impossible, and it is possible to reduce the possibility that an incorrect destination address is output.
  • the shipment sorting apparatus 200 has a low probability of being unable to output and outputs a destination address with high accuracy, and can sort the shipment into an optimum sorting box.
  • the sorting unit 250 sorts the shipment using the destination address output from the address recognition unit 240 by referring to the sorting table in which the address, the address, and the sorting box number are associated.
  • the sorting unit 250 can include a sorting table in which an ID string and a sorting box number are associated with each other.
  • the address recognition unit 240 outputs the generated ID string as it is to the sorting unit 250 without determining a destination address corresponding to the generated ID string.
  • the sorting unit 250 sorts the shipment using the ID string output from the address recognition unit 240.
  • each step of character recognition described above is described in a program format and stored in a recording medium or the like, and this program is read out from the recording medium by the CPU of the character recognition device 100 and the shipment sorter 200. Can be executed.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Character Discrimination (AREA)

Abstract

This character recognition device is provided with: a word table in which multiple words are registered in association with a specified character string; a reading means that reads image information of a display object in which the item to be recognized is displayed; a classifying means that classifies the image information into unit images; an analysis means that analyzes each of the unit images, arranges said unit images in order of the classifications thereof, and outputs said unit images as an analysis result; and a recognition means that, on the basis of the analysis result, extracts a character string from the word table, and outputs a word associated with the extracted character string, said word being output as the recognition result of the item to recognize.

Description

文字認識装置、それを備えた区分装置、文字認識方法および制御プログラムCharacter recognition device, sorting device including the same, character recognition method and control program
　本発明は、紙面などに印刷または記入された文字を読み取る文字認識装置、それを備えた区分装置、文字認識方法および制御プログラムに関する。 The present invention relates to a character recognition device that reads characters printed or written on a paper surface, a sorting device including the character recognition device, a character recognition method, and a control program.
　郵便物の区分は、一般的に、イメージスキャナ等を用いて郵便物の画像をスキャンし、ＯＣＲ（Optical Character Recognition）技術等を用いてスキャン画像から宛名や住所などの宛先情報を読み取り、読み取った宛先情報に基づいて郵便物を区分することにより行われる。 In general, mail items are scanned by scanning images of mail using an image scanner, etc., and reading destination information such as address and address from the scanned image using OCR (Optical Character Recognition) technology. This is done by sorting mail items based on destination information.
　郵便物の区分に関する技術は、例えば、特許文献１に開示されている。また、スキャン画像から宛先情報を読み取る技術は、例えば、非特許文献１に開示されている。さらに、スキャン画像に基づく宛先情報の読み取りは、スキャン画像の特徴を抽出して行うことができ、この特徴の抽出方法は、例えば、非特許文献２に開示されている。 For example, Patent Document 1 discloses a technique related to classification of mail items. A technique for reading destination information from a scanned image is disclosed in Non-Patent Document 1, for example. Furthermore, reading of destination information based on a scanned image can be performed by extracting features of the scanned image, and this feature extraction method is disclosed in Non-Patent Document 2, for example.
　ここで、スキャン画像から宛名や住所などの宛先情報を読み取る方法としては、２種類のアプローチが知られている。 Here, two types of approaches are known as methods for reading destination information such as an address and an address from a scanned image.
　一つ目は、スキャン画像を１文字毎の文字画像に分解し、分解した文字画像をそれぞれ認識するエンジンに入力して認識結果である文字コードを取得し、取得した文字コードを組み合わせて宛先情報として出力する方法である。 First, the scanned image is decomposed into character images for each character, input to the engine for recognizing each decomposed character image to acquire a character code as a recognition result, and destination information by combining the acquired character codes As the output method.
　二つ目は、スキャン画像を文字画像に分解することなくそのまま認識エンジンに入力し、スキャン画像そのものの認識結果を宛先情報として出力する方法である。 The second is a method in which the scanned image is directly input to the recognition engine without being decomposed into a character image, and the recognition result of the scanned image itself is output as destination information.
特開２００３－２１１０９３号公報JP 2003-211093 A
　郵便物のスキャン画像を１文字毎の文字画像に分解して認識する場合、認識処理が単純であると共に高い精度で文字コードを取得することができる。しかし、誤った文字コードを取得した場合は、それを修正することが出来ずに、誤った宛先情報が出力される。 When recognizing a scanned image of a mail piece by separating it into character images for each character, the recognition process is simple and a character code can be acquired with high accuracy. However, if an incorrect character code is acquired, it cannot be corrected and incorrect destination information is output.
　一方、スキャン画像を文字画像に分解することなくそのまま認識する場合、予め登録されている宛先情報のうちのどれに相当するかを決定すればよく、誤った宛先情報が出力される確率が低い。例えば、東京２３区の住所を読み取る場合、２３個の候補のうちから１個を選択できればよい。しかし、スキャン画像をそのまま認識する場合、認識処理の負荷が大きくなると共に、候補を１個に絞り込むことができない場合は認識不可となる。 On the other hand, when the scanned image is recognized as it is without being decomposed into a character image, it is only necessary to determine which of the destination information is registered in advance, and the probability that erroneous destination information is output is low. For example, when reading the address of Tokyo 23 wards, it is only necessary to select one out of 23 candidates. However, when a scanned image is recognized as it is, the recognition processing load increases, and if the candidate cannot be narrowed down to one, recognition is impossible.
　本発明の目的は、上記課題に鑑み、認識不可となる確率が低いと共に高い精度で単語を認識することができる、文字認識装置、それを備えた区分装置、文字認識方法および制御プログラムを提供することにある。 In view of the above-described problems, an object of the present invention is to provide a character recognition device, a sorting device including the character recognition device, a character recognition method, and a control program that can recognize words with high accuracy and low probability of being unrecognizable. There is.
　上記目的を達成するために本発明に係る文字認識装置は、複数の単語が所定の文字列と対応づけられて登録されている単語テーブルと、認識対象が表示されている表示体の画像情報を読み取る読取手段と、読み取った画像情報を単位画像に区分する区分手段と、区分した単位画像をそれぞれ解析して区分順に並べ、解析結果として出力する解析手段と、解析結果に基づいて単語テーブルから文字列を抽出し、抽出した文字列に対応付けられている単語を認識対象の認識結果として出力する認識手段と、を備える。 In order to achieve the above object, a character recognition device according to the present invention includes a word table in which a plurality of words are registered in association with a predetermined character string, and image information of a display body on which a recognition target is displayed. Reading means for reading, dividing means for dividing the read image information into unit images, analyzing means for analyzing the divided unit images, arranging them in the order of division, and outputting them as analysis results, and characters from the word table based on the analysis results Recognizing means for extracting a string and outputting a word associated with the extracted character string as a recognition target recognition result.
　上記目的を達成するために本発明に係る区分装置は、認識対象が表示されている発送物を搬送する搬送装置と、搬送中の発送物から画像情報を読み取り、認識対象の認識結果を出力する上記の文字認識装置と、出力された認識結果に基づいて、搬送されて来た発送物を仕分けする仕分装置と、を備える。 In order to achieve the above object, a sorting apparatus according to the present invention reads a piece of image information from a conveying device that conveys a shipment on which a recognition target is displayed, and a shipment being conveyed, and outputs a recognition result of the recognition target. The above character recognition device and a sorting device that sorts the shipments that have been conveyed based on the output recognition result.
　上記目的を達成するために本発明に係る文字認識方法は、複数の単語が所定の文字列と対応づけられて登録されている単語テーブルを用いた文字認識であって、認識対象が表示されている表示体の画像情報を読み取り、読み取った画像情報を単位画像に区分し、区分した単位画像をそれぞれ解析して区分順に並べ、解析結果として出力し、解析結果に基づいて単語テーブルから文字列を抽出し、抽出した文字列に対応付けられている単語を認識対象の認識結果として出力する。 To achieve the above object, a character recognition method according to the present invention is a character recognition using a word table in which a plurality of words are registered in association with a predetermined character string, and a recognition target is displayed. The image information of the display body is read, the read image information is divided into unit images, the divided unit images are analyzed and arranged in the order of division, and output as analysis results, and character strings are extracted from the word table based on the analysis results. The extracted word string is output as a recognition target recognition result.
　上記目的を達成するために本発明に係る制御プログラムは、複数の単語が所定の文字列と対応づけられて登録されている単語テーブルを備えた文字認識装置のコンピュータが実行する制御プログラムである。この制御プログラムは、認識対象が表示されている表示体の画像情報を読み取る機能、読み取った画像情報を単位画像に区分する機能、区分した単位画像をそれぞれ解析して区分順に並べ、解析結果として出力する機能、解析結果に基づいて単語テーブルから文字列を抽出し、抽出した文字列に対応付けられている単語を認識対象の認識結果として出力する機能、を、をコンピュータに実行させる。 In order to achieve the above object, a control program according to the present invention is a control program executed by a computer of a character recognition device having a word table in which a plurality of words are registered in association with a predetermined character string. This control program has the function of reading the image information of the display object on which the recognition target is displayed, the function of dividing the read image information into unit images, and analyzing the divided unit images and arranging them in the order of division and outputting them as analysis results. And a function of extracting a character string from the word table based on the analysis result and outputting a word associated with the extracted character string as a recognition target recognition result.
　本発明に係る文字認識装置、それを備えた区分装置、文字認識方法および制御プログラムは、認識不可となる確率が低いと共に高い精度で単語を認識することができる。 The character recognition device according to the present invention, the sorting device including the character recognition device, the character recognition method, and the control program can recognize words with high accuracy with a low probability of being unrecognizable.
本発明の第１の実施形態に係る区分装置１０の構成図である。It is a lineblock diagram of sorting device 10 concerning a 1st embodiment of the present invention. 本発明の第１の実施形態に係る文字認識装置３０のブロック構成図である。It is a block block diagram of the character recognition apparatus 30 which concerns on the 1st Embodiment of this invention. 本発明の第２の実施形態に係る文字認識装置１００のブロック構成図である。It is a block block diagram of the character recognition apparatus 100 which concerns on the 2nd Embodiment of this invention. 本発明の第２の実施形態に係る英字の分類テーブル１５０の一例である。It is an example of the alphabetic classification table 150 which concerns on the 2nd Embodiment of this invention. 本発明の第２の実施形態に係るタイ文字の分類テーブル１５０の一例である。It is an example of the Thai character classification table 150 according to the second embodiment of the present invention. 本発明の第２の実施形態に係る変換テーブル１７０の一例である。It is an example of the conversion table 170 concerning the 2nd Embodiment of this invention. 本発明の第２の実施形態に係る文字認識装置１００の動作フロー図である。It is an operation | movement flowchart of the character recognition apparatus 100 which concerns on the 2nd Embodiment of this invention. 本発明の第３の実施形態に係る発送物区分機２００の構成図である。It is a block diagram of the dispatch sorter 200 which concerns on the 3rd Embodiment of this invention. 本発明の第３の実施形態に係る仕分けテーブルの一例である。It is an example of the sorting table which concerns on the 3rd Embodiment of this invention.
　（第１の実施形態）
　第１の実施形態について説明する。本実施形態では、文字認識装置を備えた区分装置を適用する。本実施形態に係る区分装置の構成図を図１Ａに、文字認識装置のブロック構成図を図１Ｂに示す。 (First embodiment)
A first embodiment will be described. In this embodiment, a sorting device provided with a character recognition device is applied. FIG. 1A shows a configuration diagram of the sorting apparatus according to the present embodiment, and FIG. 1B shows a block configuration diagram of the character recognition device.
　図１Ａにおいて、本実施形態に係る区分装置１０は、搬送装置２０、文字認識装置３０および仕分装置４０を備える。 1A, the sorting apparatus 10 according to the present embodiment includes a transport device 20, a character recognition device 30, and a sorting device 40.
　搬送装置２０は、認識対象である宛先情報が表示されている発送物５０（表示体）を搬送する。文字認識装置３０は、搬送中の発送物５０の画像情報をスキャン等によって取り込み、取り込んだ画像情報から宛先情報を認識し、認識した宛先情報を仕分装置４０へ出力する。仕分装置４０は、文字認識装置３０から出力された宛先情報に基づいて、搬送されて来た発送物５０を仕分けする。 The transport device 20 transports the shipment 50 (display body) on which the destination information that is the recognition target is displayed. The character recognition device 30 captures image information of the shipment 50 being conveyed by scanning or the like, recognizes destination information from the captured image information, and outputs the recognized destination information to the sorting device 40. The sorting device 40 sorts the shipped items 50 that have been transported based on the destination information output from the character recognition device 30.
　文字認識装置３０について詳細に説明する。図１Ｂに示すように、文字認識装置３０は、単語テーブル３１、読取手段３２、区分手段３３、解析手段３４および認識手段３５を備える。 The character recognition device 30 will be described in detail. As shown in FIG. 1B, the character recognition device 30 includes a word table 31, a reading unit 32, a sorting unit 33, an analysis unit 34, and a recognition unit 35.
　単語テーブル３１には、複数の単語が所定の文字列と対応づけられて登録されている。本実施形態において、単語テーブル３１には、発送物５０を区分けに必要な全ての宛先情報が単語として登録され、該宛先情報にそれぞれ、宛先情報を抽出するための文字列が紐付けられている。 In the word table 31, a plurality of words are registered in association with a predetermined character string. In the present embodiment, in the word table 31, all destination information necessary to classify the shipment 50 is registered as words, and a character string for extracting the destination information is associated with each destination information. .
　読取手段３２は、認識対象が表示されている発送物５０の画像情報を読み取る。読取手段３２としては、例えば、スキャナを適用することができる。 The reading means 32 reads the image information of the shipment 50 on which the recognition target is displayed. As the reading unit 32, for example, a scanner can be applied.
　区分手段３３は、読取手段３２が発送物５０から読み取った画像情報を１文字毎の文字画像（単位画像）に区分し、画像情報を形成していた順番（区分順）を付加して解析手段３４へ出力する。 The sorting unit 33 sorts the image information read by the reading unit 32 from the shipment 50 into character images (unit images) for each character, and adds the order (classification order) in which the image information was formed to analyze the image information. 34.
　解析手段３４は、区分手段３３から出力された文字画像の特徴量をそれぞれ演算し、演算した特徴量を有する文字をそれぞれ特定する。解析手段３４はさらに、特定した文字を区分順に並べて文字列を生成し、解析結果として認識手段３５へ出力する。 The analyzing unit 34 calculates the feature amount of the character image output from the sorting unit 33 and specifies each character having the calculated feature amount. The analysis unit 34 further generates a character string by arranging the specified characters in the order of division, and outputs the character string to the recognition unit 35 as an analysis result.
　認識手段３５は、解析手段３４から出力された解析結果と一致する文字列を単語テーブル３１から抽出し、抽出した文字列に対応付けられている宛先情報（単語）を認識結果として出力する。 The recognition unit 35 extracts a character string that matches the analysis result output from the analysis unit 34 from the word table 31, and outputs destination information (word) associated with the extracted character string as a recognition result.
　以上のように構成された文字認識装置３０および区分装置１０は、発送物５０の画像情報を所定の文字画像（単位画像）に区分し、文字画像をそれぞれ解析する。この場合、解析処理が単純であると共に高い精度で文字を特定することができる。 The character recognition device 30 and the sorting device 10 configured as described above classify the image information of the shipment 50 into predetermined character images (unit images), and analyze the character images respectively. In this case, the analysis process is simple and a character can be specified with high accuracy.
　また、本実施形態に係る文字認識装置３０および区分装置１０は、複数の宛先情報（単語）が文字列と対応づけられて登録されている単語テーブル３１を備え、区分順に出力された文字（解析結果）と一致する文字列を単語テーブル３１から抽出し、抽出した文字列に対応付けられている宛先情報（単語）を認識結果として出力する。この場合、単語認識処理を、有限の文字列候補のうちから１つを抽出する処理に置き換えることができ、認識不可となる可能性を低減できると共に誤った宛先情報（単語）が認識結果として出力されることを低減できる。 In addition, the character recognition device 30 and the sorting device 10 according to this embodiment include a word table 31 in which a plurality of pieces of destination information (words) are registered in association with character strings, and characters (analysis) output in the order of sorting. A character string that matches (result) is extracted from the word table 31, and destination information (word) associated with the extracted character string is output as a recognition result. In this case, the word recognition process can be replaced with a process of extracting one of the finite character string candidates, which can reduce the possibility of being unrecognizable and output incorrect destination information (word) as a recognition result. Can be reduced.
　従って、本実施形態に係る文字認識装置３０および区分装置１０は、認識不可となる確率が低いと共に高い精度で認識対象を認識することができ、認識対象が表示されている発送物５０を適切に仕分することができる。 Therefore, the character recognition device 30 and the sorting device 10 according to the present embodiment have a low probability of being unrecognizable and can recognize the recognition target with high accuracy, and appropriately select the shipment 50 on which the recognition target is displayed. Can be sorted.
　ここで、文字認識装置３０は、解析結果と一致する文字列が単語テーブル３１に登録されていない場合、解析結果に最も近い文字列を抽出する。最も近い文字列の抽出には、編集距離を用いることができる。なお、編集距離を用いた文字列の抽出については、第２の実施形態で詳細に説明する。 Here, when the character string that matches the analysis result is not registered in the word table 31, the character recognition device 30 extracts the character string closest to the analysis result. The edit distance can be used to extract the closest character string. The extraction of the character string using the edit distance will be described in detail in the second embodiment.
　（第２の実施形態）
　第２の実施形態について説明する。本実施形態に係る文字認識装置のブロック構成図を図２に示す。図２において、データの流れを実線で、制御信号の流れを点線で示す。図２に示すように、本実施形態に係る文字認識装置１００は、入力モジュール１１０、文字切り出しモジュール１２０、特徴抽出モジュール１３０、文字分類モジュール１４０、分類テーブル１５０、変換モジュール１６０、変換テーブル１７０、出力モジュール１８０および制御モジュール１９０を備える。 (Second Embodiment)
A second embodiment will be described. FIG. 2 shows a block diagram of the character recognition device according to this embodiment. In FIG. 2, the data flow is indicated by a solid line, and the control signal flow is indicated by a dotted line. As shown in FIG. 2, the character recognition device 100 according to the present embodiment includes an input module 110, a character segmentation module 120, a feature extraction module 130, a character classification module 140, a classification table 150, a conversion module 160, a conversion table 170, and an output. A module 180 and a control module 190 are provided.
　入力モジュール１１０は、例えば、ユーザが認識対象が印刷されている誌面を画像読取台にセットしてスタートボタンを押下する等した時、誌面から単語画像を取得し、取得した単語画像を単語画像データとして文字切り出しモジュール１２０へ出力する。ここで、入力モジュール１１０としては、スキャナ等を適用することができる。入力モジュール１１０は、例えば、誌面の特定領域内をスキャン等することによって単語画像を取得し、「Ａｂｃ」の単語画像データを文字切り出しモジュール１２０へ出力する。 The input module 110 obtains a word image from the magazine, for example, when the user sets a magazine on which the recognition target is printed on the image reading stand and presses the start button, and the obtained word image is converted into the word image data. Is output to the character segmentation module 120. Here, as the input module 110, a scanner or the like can be applied. The input module 110 acquires a word image, for example, by scanning a specific area of a magazine, and outputs the word image data “Abc” to the character segmentation module 120.
　文字切り出しモジュール１２０は、入力モジュール１１０から出力された単語画像データを、１文字毎の文字画像データに分解して特徴抽出モジュール１３０へ出力する。文字切り出しモジュール１２０は、例えば、入力モジュール１１０から出力された「Ａｂｃ」の単語画像データを１文字毎の文字画像データに分解し、「Ａ」、「ｂ」および「ｃ」の３つの文字画像データを出力する。 The character cutout module 120 decomposes the word image data output from the input module 110 into character image data for each character and outputs the character image data to the feature extraction module 130. The character segmentation module 120, for example, decomposes the word image data “Abc” output from the input module 110 into character image data for each character, and three character images “A”, “b”, and “c”. Output data.
　特徴抽出モジュール１３０は、文字切り出しモジュール１２０から出力された文字画像データをそれぞれ画像処理し、各文字画像データの特徴量を抽出することにより、文字要素を決定して文字分類モジュール１４０へ出力する。特徴抽出モジュール１３０は、光学文字認識装置（ＯＣＲ）における画像処理で一般的に用いられている特徴量の抽出手法を用いることができる。また、特徴量の抽出手法において、例えば、背景技術に示した非特許文献２に開示されている、局所領域毎の文字のエッジの方向情報等を用いることができる。特徴抽出モジュール１３０は、例えば、「Ａ」、「ｂ」および「ｃ」の３つの文字画像データをそれぞれ画像処理し、「文字要素Ａ」、「文字要素ｂ」および「文字要素ｃ」を決定して文字分類モジュール１４０へ出力する。 The feature extraction module 130 performs image processing on the character image data output from the character cutout module 120, extracts the feature amount of each character image data, determines character elements, and outputs them to the character classification module 140. The feature extraction module 130 can use a feature amount extraction method generally used in image processing in an optical character recognition device (OCR). In the feature amount extraction method, for example, the direction information of the edge of the character for each local region disclosed in Non-Patent Document 2 shown in the background art can be used. The feature extraction module 130 processes, for example, three character image data “A”, “b”, and “c”, respectively, and determines “character element A”, “character element b”, and “character element c”. And output to the character classification module 140.
　文字分類モジュール１４０は、分類テーブル１５０を参照することにより、特徴抽出モジュール１３０から出力された文字要素が属する文字グループを特定する。さらに、文字分類モジュール１４０は、特定した文字グループのグループＩＤを出力順に並べてＩＤ列を生成し、生成したＩＤ列を変換モジュール１６０へ出力する。 The character classification module 140 refers to the classification table 150 to identify the character group to which the character element output from the feature extraction module 130 belongs. Further, the character classification module 140 generates an ID string by arranging the group IDs of the identified character groups in the order of output, and outputs the generated ID string to the conversion module 160.
　分類テーブル１５０には、認識対象となる単語を構成する全ての文字要素が分類されて登録されている。分類テーブル１５０においては、特徴量が近い文字要素をグループ化し、グループにそれぞれグループＩＤを付与した。なお、特徴量が近くても、それらを区別する必要がある場合には、別のグループとすることが望ましい。なお、分類テーブル１５０が請求項の文字テーブルに相当する。 In the classification table 150, all the character elements constituting the word to be recognized are classified and registered. In the classification table 150, character elements having similar feature amounts are grouped, and a group ID is assigned to each group. Note that even if feature quantities are close to each other, if it is necessary to distinguish them, it is desirable to make them separate groups. The classification table 150 corresponds to a character table in the claims.
　分類テーブル１５０の一例を図３Ａ、図３Ｂに示す。図３Ａは英字の分類テーブル１５０、図３Ｂはタイ文字の分類テーブル１５０である。図３Ａ、図３Ｂにおいて、特徴量が近い英文字およびタイ文字同士がグループ化され、各文字グループにグループＩＤが付与されている。例えば、図３Ａにおいて、特徴量が近い「ｃ」、「Ｃ」および「Ｇ」の３つの文字要素がグループ化され、グループＩＤ「３」が付与されている。なお、図３Ａ、図３Ｂに示した分類テーブル１５０では、英語およびタイ語を構成する全ての文字および数字の文字要素を登録したが、後述する変換テーブル１７０に登録されている単語に使用されていない文字および数字の文字要素は分類テーブル１５０に登録しなくてもよい。 An example of the classification table 150 is shown in FIGS. 3A and 3B. FIG. 3A shows an English character classification table 150, and FIG. 3B shows a Thai character classification table 150. 3A and 3B, English characters and Thai characters having similar feature amounts are grouped, and a group ID is assigned to each character group. For example, in FIG. 3A, three character elements “c”, “C”, and “G” having similar feature amounts are grouped and given a group ID “3”. In the classification table 150 shown in FIG. 3A and FIG. 3B, all the characters and numeric character elements constituting English and Thai are registered, but they are used for the words registered in the conversion table 170 described later. Non-letter and numeric character elements may not be registered in the classification table 150.
　文字分類モジュール１４０は、例えば、分類テーブル１５０を参照することにより、特徴抽出モジュール１３０から出力された「文字要素Ａ」が属する文字グループを特定し、特定した文字グループのグループＩＤ「２５」を取得する。また、文字分類モジュール１４０は、「文字要素ｂ」が属する文字グループのグループＩＤ「２」、「文字要素ｃ」が属する文字グループのグループＩＤ「３」を取得する。そして、文字分類モジュール１４０は、取得したグループＩＤを出力順に並べることによりＩＤ列「２５－２－３」を生成する。 The character classification module 140 identifies, for example, the character group to which the “character element A” output from the feature extraction module 130 belongs by referring to the classification table 150, and acquires the group ID “25” of the identified character group. To do. Further, the character classification module 140 acquires the group ID “2” of the character group to which “character element b” belongs, and the group ID “3” of the character group to which “character element c” belongs. Then, the character classification module 140 generates the ID string “25-2-3” by arranging the acquired group IDs in the order of output.
　変換モジュール１６０は、変換テーブル１７０を参照することにより、文字分類モジュール１４０から出力されたＩＤ列に対応する単語を決定し、決定した単語を出力モジュール１８０へ出力する。 The conversion module 160 refers to the conversion table 170 to determine a word corresponding to the ID string output from the character classification module 140, and outputs the determined word to the output module 180.
　変換テーブル１７０には、複数の単語がＩＤ列と紐付けられて登録されている。なお、変換テーブル１７０が請求項の単語テーブルに相当する。変換テーブル１７０の一例を図４に示す。図４において、例えば、単語「Ａｂｃ　Ｒｏａｄ」にはＩＤ列「２５－２－３　３６－１４－１－２７」が紐付けられている。 In the conversion table 170, a plurality of words are registered in association with the ID string. The conversion table 170 corresponds to the word table in the claims. An example of the conversion table 170 is shown in FIG. In FIG. 4, for example, an ID string “25-2-3 36-14-1-27” is associated with the word “Abc Road”.
　変換モジュール１６０は、例えば、文字分類モジュール１４０から出力されたＩＤ列「２５－２－３　３６－１４－１－２７」と一致するＩＤ列を変換テーブル１７０から抽出し、抽出したＩＤ列に紐付けられている単語「Ａｂｃ　Ｒｏａｄ」を出力モジュール１８０へ出力する。 For example, the conversion module 160 extracts an ID string that matches the ID string “25-2-3 36-14-1-27” output from the character classification module 140 from the conversion table 170, and associates the ID string with the extracted ID string. The attached word “Abc Load” is output to the output module 180.
　さらに、本実施形態に係る変換モジュール１６０は、文字分類モジュール１４０から出力されたＩＤ列と一致するＩＤ列が変換テーブル１７０に登録されていない場合、最も近いＩＤ列を変換テーブル１７０から抽出する。例えば、「Ａｂｃ　Ｒｏａｄ」の文字認識において、文字分類モジュール１４０が「Ｒｏａｄ」の第２番目の「ｏ」を「ｅ」と誤認識した場合、文字分類モジュール１４０からはＩＤ列「２５－２－３　３６－１４－１－２７」ではなくＩＤ列「２５－２－３　３６－５－１－２７」が出力される。この場合、変換テーブル１７０にはＩＤ列「２５－２－３　３６－５－１－２７」は登録されていないため、変換モジュール１６０はＩＤ列「２５－２－３　３６－５－１－２７」と紐付けられている単語を抽出することができない。 Furthermore, the conversion module 160 according to the present embodiment extracts the closest ID string from the conversion table 170 when an ID string that matches the ID string output from the character classification module 140 is not registered in the conversion table 170. For example, in the character recognition of “Abc Road”, if the character classification module 140 misrecognizes the second “o” of “Load” as “e”, the character classification module 140 receives the ID string “25-2- ID string “25-2-3 36-5-1-27” is output instead of “3 36-14-27”. In this case, since the ID column “25-2-3 36-5-1-27” is not registered in the conversion table 170, the conversion module 160 uses the ID column “25-2-3 36-5-1-27”. ”Cannot be extracted.
　本実施形態に係る変換モジュール１６０は、文字分類モジュール１４０から出力されたＩＤ列が変換テーブル１７０に登録されていない場合、距離尺度を用いて文字分類モジュール１４０から出力されたＩＤ列に最も近いＩＤ列を変換テーブル１７０から抽出する。すなわち、ＩＤ列を文字列と考えると、文字列同士の近さは一方の文字列からもう一方の文字列に変換する際のコストとして定義することができる。ＩＤ列を構成する文字の挿入、文字の削除、文字の入れ替えに対してそれぞれコストを定義し、変換に要したコストの総計をＩＤ列間の距離と考え、変換コストが最も小さいＩＤ列、すなわち、ＩＤ列間距離が最も小さいＩＤ列を対応するＩＤ列とする。 When the ID string output from the character classification module 140 is not registered in the conversion table 170, the conversion module 160 according to the present embodiment uses the distance scale to obtain the ID closest to the ID string output from the character classification module 140. A column is extracted from the conversion table 170. That is, when the ID string is considered as a character string, the closeness between the character strings can be defined as a cost for converting from one character string to the other character string. Costs are defined for insertion, deletion, and replacement of characters constituting the ID string, and the total cost required for conversion is considered as the distance between the ID strings. The ID string having the smallest distance between ID strings is taken as the corresponding ID string.
　この考え方は、編集距離（レーベンシュタイン距離、エディットディスタンス）として知られており、挿入・削除・入れ替えに対するそれぞれのコストを１とする場合、上述のＩＤ列「２５－２－３　３６－１４－１－２７」とＩＤ列「２５－２－３　３６－５－１－２７」とのＩＤ列間距離は１文字の入れ替えなので１となる。なお、所定の閾値を設定し、閾値より小さいＩＤ列間距離を有するＩＤ列が見つからない場合には、認識結果なし（棄却）とすることもできる。 This concept is known as edit distance (Levenstein distance, edit distance). When each cost for insertion / deletion / replacement is 1, the ID string “25-2-3 36-14-1” described above is used. The distance between ID columns of “−27” and the ID column “25-2-3 36-5-1-27” is 1 because one character is replaced. Note that if a predetermined threshold value is set and an ID string having an ID string distance smaller than the threshold value is not found, no recognition result can be set (rejected).
　出力モジュール１８０は、変換モジュール１６０から出力された単語「Ａｂｃ　Ｒｏａｄ」を、誌面に印刷されている認識対象の認識結果として出力する。 The output module 180 outputs the word “Abc Load” output from the conversion module 160 as the recognition result of the recognition target printed on the magazine.
　制御モジュール１９０は、入力モジュール１１０、文字切り出しモジュール１２０、特徴抽出モジュール１３０、分類モジュール１０４、変換モジュール１６０および出力モジュール１８０の動作をそれぞれ制御する。 The control module 190 controls operations of the input module 110, the character segmentation module 120, the feature extraction module 130, the classification module 104, the conversion module 160, and the output module 180, respectively.
　上述の各モジュールは、入力部、処理部、記憶部および出力部を有し、デジタルデータを処理することによって上記の動作を行う一般的なコンピュータや、誌面をスキャンしてデジタル画像データを生成するスキャナ等により、実現することができる。 Each of the modules described above has an input unit, a processing unit, a storage unit, and an output unit, and generates digital image data by scanning a magazine or a general computer that performs the above operation by processing digital data. It can be realized by a scanner or the like.
　次に、本実施形態に係る文字認識装置１００の動作フローを説明する。本実施形態に係る文字認識装置１００の動作フローを図５に示す。 Next, the operation flow of the character recognition device 100 according to this embodiment will be described. FIG. 5 shows an operation flow of the character recognition apparatus 100 according to this embodiment.
　図５において、ユーザが誌面を文字認識装置１００の画像読取台等にセットしてスタートボタンを押下する等することにより、入力モジュール１１０は、誌面の特定領域内をスキャンし、単語画像を取得する。入力モジュール１１０は、取得した単語画像を単語画像データとして文字切り出しモジュール１２０へ出力する（ステップＳ１０１）。 In FIG. 5, when the user sets the magazine on the image reading table of the character recognition device 100 and presses the start button, the input module 110 scans a specific area of the magazine and acquires a word image. . The input module 110 outputs the acquired word image as word image data to the character segmentation module 120 (step S101).
　文字切り出しモジュール１２０は、入力モジュール１１０から出力された単語画像データを、１文字毎の文字画像データに分解して特徴抽出モジュール１３０へ出力する。（ステップＳ１０２）。１文字毎の文字画像データに分割する方法としては、例えば、単語画像データについて、文字列の方向に沿って画素値を投影してヒストグラム作成し、ヒストグラムの谷と谷の間を１文字とする方法がある。または、画素の連結領域を調べ、互いに隣接する連結領域を統合することによって１文字毎の画像を作成するラベリング法を適用することもできる。 The character cutout module 120 decomposes the word image data output from the input module 110 into character image data for each character and outputs the character image data to the feature extraction module 130. (Step S102). As a method of dividing character image data for each character, for example, for word image data, a histogram is created by projecting pixel values along the direction of the character string, and one character is defined between the valleys of the histogram. There is a way. Alternatively, it is possible to apply a labeling method in which an image for each character is created by examining connected regions of pixels and integrating adjacent connected regions.
　特徴抽出モジュール１３０は、文字切り出しモジュール１２０から出力された文字画像データをそれぞれ画像処理し、各文字画像データの特徴量を抽出することにより決定した文字要素を文字分類モジュール１４０へ出力する（ステップＳ１０３）。 The feature extraction module 130 performs image processing on the character image data output from the character cutout module 120, and outputs the character elements determined by extracting the feature amount of each character image data to the character classification module 140 (step S103). ).
　文字分類モジュール１４０は、分類テーブル１５０を参照し、特徴抽出モジュール１３０から出力された文字要素が属する文字グループをそれぞれ特定し、特定した文字グループのグループＩＤを取得する。文字分類モジュール１４０は、取得したグループＩＤを特徴量の出力順に並べて、ＩＤ列として変換モジュール１６０へ出力する（ステップＳ１０４）。 The character classification module 140 refers to the classification table 150, specifies each character group to which the character element output from the feature extraction module 130 belongs, and acquires the group ID of the specified character group. The character classification module 140 arranges the acquired group IDs in the output order of the feature values, and outputs them to the conversion module 160 as an ID string (step S104).
　変換モジュール１６０は、変換テーブル１７０を参照して、文字分類モジュール１４０から出力されたＩＤ列に対応付けられている単語を決定し、決定した単語を出力モジュール１８０へ出力する（ステップＳ１０５）。なお、文字分類モジュール１４０から出力されたＩＤ列が変換テーブル１７０に存在しない場合、変換モジュール１６０は、距離尺度を用いて文字分類モジュール１４０から出力されたＩＤ列に最も近いＩＤ列を変換テーブル１７０から抽出し、抽出したＩＤ列に対応付けられている単語を出力する。 The conversion module 160 refers to the conversion table 170, determines a word associated with the ID string output from the character classification module 140, and outputs the determined word to the output module 180 (step S105). When the ID string output from the character classification module 140 does not exist in the conversion table 170, the conversion module 160 uses the distance scale to select the ID string closest to the ID string output from the character classification module 140 as the conversion table 170. And the word associated with the extracted ID string is output.
　出力モジュール１８０は、変換モジュール１６０から出力された単語を、誌面の特定領域に記載されている単語画像に対する認識結果として出力する（ステップＳ１０６）。 The output module 180 outputs the word output from the conversion module 160 as a recognition result for the word image described in the specific area of the magazine (step S106).
　以上のように、本実施形態に係る文字認識装置１００は、単語画像データを１文字毎の文字画像データに分解し、文字画像データごとに属する文字グループを特定する。文字画像データを１つずつ認識する場合、認識処理が単純であると共に高い精度で属する文字グループを特定することができる。 As described above, the character recognition device 100 according to the present embodiment decomposes the word image data into character image data for each character, and specifies a character group belonging to each character image data. When character image data is recognized one by one, the recognition process is simple and a character group to which the character group belongs can be specified with high accuracy.
　また、本実施形態に係る文字認識装置１００は、複数の単語がＩＤ列と紐付けられて登録されている変換テーブル１７０を備え、特定した文字グループのグループＩＤによって構成されるＩＤ列が、変換テーブル１７０に登録されている単語のうちのどれに対応するかを決定する。この場合、単語認識処理を有限の候補のうちから１つを選択する選択処理に置き換えることができ、認識不可となる可能性を低減できると共に誤った単語認識結果が出力される可能性を低減できる。また、有限の候補のうちから１つを選択する選択処理は単語認識処理より演算負荷が小さいことから、処理速度を向上させることができる。 In addition, the character recognition device 100 according to the present embodiment includes a conversion table 170 in which a plurality of words are registered in association with an ID string, and an ID string configured by a group ID of the specified character group is converted. It is determined which of the words registered in the table 170 corresponds to. In this case, the word recognition process can be replaced with a selection process for selecting one of the finite candidates, so that the possibility of being unable to recognize can be reduced and the possibility of outputting an incorrect word recognition result can be reduced. . In addition, the selection process for selecting one of the finite candidates has a smaller calculation load than the word recognition process, so that the processing speed can be improved.
　従って、本実施形態に係る文字認識装置１００は、認識不可となる確率が低いと共に高い精度で単語を認識することができる。 Therefore, the character recognition device 100 according to the present embodiment can recognize words with high accuracy and a low probability of being unrecognizable.
　なお、分類テーブル１５０には、変換テーブル１７０に登録されている単語を構成する文字および数字のみを登録しておけば良い。この場合、文字分類モジュール１４０の演算処理を必要最小限にすることができ、単語認識の処理速度をさらに向上させることができる。 In the classification table 150, only letters and numbers constituting the words registered in the conversion table 170 may be registered. In this case, the arithmetic processing of the character classification module 140 can be minimized, and the word recognition processing speed can be further improved.
　また、本実施形態に係る分類テーブル１５０では、区別する必要性が低い類似文字要素同士をグループ化した。この場合、認識対象の単語画像の中に類似文字が含まれていても、類似文字に影響を受けることなく単語認識を行うことができる。例えば、誌面に記載されている単語画像が「ＡｂＣ」であった場合でも、分類テーブル１５０において「Ｃ」と「ｃ」とが同じグループに属していることにより、変換テーブル１７０を参照し、「Ａｂｃ」の正しい単語認識結果が出力される。なお、変換テーブル１７０に「ＡｂＣ」および「Ａｂｃ」の両方が登録されている場合、すなわち、両者を区別する必要がある場合には、分類テーブル１５０において「Ｃ」と「ｃ」とをそれぞれ異なるグループに登録しておく。 Further, in the classification table 150 according to the present embodiment, similar character elements that are less necessary to be distinguished are grouped. In this case, even if similar characters are included in the word image to be recognized, word recognition can be performed without being affected by the similar characters. For example, even when the word image described in the magazine is “AbC”, the “C” and “c” in the classification table 150 belong to the same group, so that the conversion table 170 is referred to, and “ The correct word recognition result of “Abc” is output. When both “AbC” and “Abc” are registered in the conversion table 170, that is, when it is necessary to distinguish between both, “C” and “c” are different from each other in the classification table 150. Register with the group.
　ここで、図３に示した分類テーブル１５０においては、１つの文字グループに複数の文字要素を登録したが、これに限定されない。例えば、文字グループの代表的な文字要素や、文字グループに属する文字要素の特徴量の平均等を登録しておくこともできる。また、文字グループに属する文字要素の特徴量を主成分分析し、主成分を合成した主軸（固有ベクトル）を登録しておくこともできる。この場合、文字分類モジュール１４０は、特徴抽出モジュール１３０から出力された文字画像データの特徴量を、分類テーブル１５０に登録されている文字グループの文字要素の特徴量と順次比較し、最も近い特長量を有する文字グループを特定する。 Here, in the classification table 150 shown in FIG. 3, a plurality of character elements are registered in one character group, but the present invention is not limited to this. For example, a representative character element of a character group, an average of feature amounts of character elements belonging to the character group, and the like can be registered. It is also possible to register principal axes (eigenvectors) obtained by performing principal component analysis on the feature amounts of character elements belonging to a character group and combining the principal components. In this case, the character classification module 140 sequentially compares the feature amount of the character image data output from the feature extraction module 130 with the feature amount of the character element of the character group registered in the classification table 150, and the closest feature amount. Identify character groups that have
　（第３の実施形態）
　第３の実施形態について説明する。本実施形態では、発送物区分機を適用する。本実施形態に係る発送物区分機の構成図を図６に示す。また、図６にデータおよび制御信号の流れを実線で、発送物の流れを破線で示す。図６に示すように、本実施形態に係る発送物区分機２００は、供給部２１０、搬送部２２０、スキャナ部２３０、住所認識部２４０、仕分部２５０および制御部２６０を備える。 (Third embodiment)
A third embodiment will be described. In this embodiment, a shipment sorting machine is applied. FIG. 6 shows a configuration diagram of the shipment sorting machine according to the present embodiment. In FIG. 6, the flow of data and control signals is indicated by a solid line, and the flow of shipments is indicated by a broken line. As shown in FIG. 6, the shipment sorting machine 200 according to the present embodiment includes a supply unit 210, a transport unit 220, a scanner unit 230, an address recognition unit 240, a sorting unit 250, and a control unit 260.
　供給部２１０は、宛先住所が表示されている発送物を１通ずつ、搬送部２２０へ供給する。搬送部２２０は、供給部２１０から供給された発送物を、仕分部２５０まで搬送する。スキャナ部２３０は、搬送部２２０によって搬送されている発送物の宛先住所が表示されている領域の画像を撮像し、取得した撮像画像を住所認識部２４０へ出力する。 The supply unit 210 supplies the shipments on which the destination addresses are displayed one by one to the transport unit 220. The transport unit 220 transports the shipment supplied from the supply unit 210 to the sorting unit 250. The scanner unit 230 captures an image of an area where the destination address of the shipment being transported by the transport unit 220 is displayed, and outputs the acquired captured image to the address recognition unit 240.
　住所認識部２４０は、図示しない分類テーブルおよび変換テーブルを備え、分類テーブルおよび変換テーブルを参照して、スキャナ部２３０から出力された撮像画像から宛先住所を認識し、認識した宛先住所を仕分部２５０へ出力する。 The address recognition unit 240 includes a classification table and a conversion table (not shown), refers to the classification table and the conversion table, recognizes the destination address from the captured image output from the scanner unit 230, and sorts the recognized destination address into the sorting unit 250. Output to.
　住所認識部２４０は、スキャナ部２３０から出力された撮像画像から住所領域を判別し、判別した住所領域の撮像画像を単語画像データとして取得する機能、取得した単語画像データを住所を構成している住所単語に分解する機能、および、第２の実施形態で説明した文字認識装置１００の機能を有する。なお、発送物の撮像画像から住所単語を取得する方法は、例えば、背景技術に示した非特許文献１に開示されている方法を適用することができる。 The address recognition unit 240 determines an address area from the captured image output from the scanner unit 230, acquires a captured image of the determined address area as word image data, and configures the acquired word image data as an address. It has the function of decomposing into address words and the function of the character recognition device 100 described in the second embodiment. In addition, the method currently disclosed by the nonpatent literature 1 shown to background art is applicable to the method of acquiring an address word from the captured image of a shipment thing, for example.
　具体的には、住所認識部２４０は、スキャナ部２３０から出力された発送物の撮像画像から住所単語を取得し、単語画像データに分割する。住所認識部２４０は、分割した単語画像データの特徴量をそれぞれ演算し、演算した特徴量に基づいて決定した文字要素が属する文字グループを特定してＩＤ列を生成し、生成したＩＤ列に対応する宛先住所を変換テーブルから選択して出力する。 Specifically, the address recognition unit 240 acquires an address word from the captured image of the shipment output from the scanner unit 230 and divides it into word image data. The address recognition unit 240 calculates the feature amount of each divided word image data, identifies the character group to which the character element determined based on the calculated feature amount belongs, generates an ID string, and corresponds to the generated ID string The destination address to be selected is selected from the conversion table and output.
　仕分部２５０は、宛先住所と区分箱番号とが関連づけられて登録されている仕分けテーブルを備える。仕分部２５０は、住所認識部２４０から出力された宛先住所に関連付けられている区分箱番号を仕分けテーブルから抽出し、搬送部２２０によって搬送されてきた発送物を、抽出した区分箱番号によって特定される区分箱に仕分けする。なお、仕分けテーブルの一例を図７に示す。図７に示した仕分けテーブルには、住所、番地および区分箱番号がそれぞれ関連付けられて登録されている。 The sorting unit 250 includes a sorting table in which a destination address and a sorting box number are registered in association with each other. The sorting unit 250 extracts the sorting box number associated with the destination address output from the address recognition unit 240 from the sorting table, and identifies the shipment that has been transported by the transporting unit 220 by the extracted sorting box number. Sort into separate boxes. An example of the sorting table is shown in FIG. In the sorting table shown in FIG. 7, an address, an address, and a sorting box number are associated and registered.
　制御部２６０は、供給部２１０、搬送部２２０、スキャナ部２３０、住所認識部２４０および仕分部２５０の動作をそれぞれ制御する。 The control unit 260 controls the operations of the supply unit 210, the transport unit 220, the scanner unit 230, the address recognition unit 240, and the sorting unit 250, respectively.
　発送物区分機２００を構成する供給部２１０、搬送部２２０および仕分部２５０は、発送物をハンドリングするための機械的な機構であり、スキャナ部２３０は、発送物の画像を撮像する光電変換機構である。これらの機構は、例えば、背景技術に示した特許文献１に開示されている郵便区分機などの一般的な装置で実現することができる。 The supply unit 210, the conveyance unit 220, and the sorting unit 250 constituting the shipment sorter 200 are mechanical mechanisms for handling the shipment, and the scanner unit 230 is a photoelectric conversion mechanism that captures an image of the shipment. It is. These mechanisms can be realized by a general apparatus such as a mail sorting machine disclosed in Patent Document 1 shown in the background art.
　本実施形態に係る発送物区分機２００を用いて東京２３区の区名を認識する場合について説明する。 A case will be described in which the ward name of Tokyo 23 wards is recognized using the shipment sorting machine 200 according to the present embodiment.
　先ず、供給部２１０は、発送物の宛先住所が表示されている領域がスキャナ部２３０側に向くようにして、発送物を１通ずつ搬送部２２０へ供給する。スキャナ部２３０は、発送物が搬送部２２０によって仕分部２５０に搬送される間に発送物の宛先住所が表示されている領域の撮像画像を取得し、住所認識部２４０へ出力する。 First, the supply unit 210 supplies the shipments one by one to the transport unit 220 so that the area where the destination address of the shipment is displayed faces the scanner unit 230 side. The scanner unit 230 acquires a captured image of an area where the destination address of the shipment is displayed while the shipment is being conveyed by the conveyance unit 220 to the sorting unit 250, and outputs the captured image to the address recognition unit 240.
　住所認識部２４０は、撮像画像を文字画像データに分解し、文字画像データごとに特徴量を演算する。住所認識部２４０は、「大田区」、「港区」、「中央区」等の２３個の区名で使用されている全ての文字（３９文字）およびこの３９文字に関する類似文字がグループ化されて登録されている分類テーブルを備え、この分類テーブルを参照して、演算した特徴量に基づいて決定した文字要素が属する文字グループを特定する。 The address recognition unit 240 decomposes the captured image into character image data, and calculates a feature amount for each character image data. The address recognition unit 240 groups all characters (39 characters) used in 23 ward names such as “Ota Ward”, “Minato Ward”, “Chuo Ward”, and similar characters related to these 39 characters. The character group to which the character element determined based on the calculated feature amount belongs is specified with reference to the classification table.
　住所認識部２４０は、さらに、特定した文字グループのグループＩＤを表示順に並べてＩＤ列を生成し、２３個の区名とＩＤ列とが関連づけられて登録されている変換テーブルを参照して、ＩＤ列に対応づけられている宛先住所を仕分部２５０へ出力する。変換テーブルには、ＩＤ列と対応付けられて「大田区」が登録されているため、「太田区」と誤植されている発送物の宛先住所が「大田区」と正しく認識される。 The address recognition unit 240 further generates an ID string by arranging the group IDs of the identified character groups in the display order, and refers to the conversion table in which 23 ward names and the ID string are associated and registered. The destination address associated with the column is output to the sorting unit 250. Since “Ota Ward” is registered in the conversion table in association with the ID column, the destination address of the shipment that is misprinted as “Ota Ward” is correctly recognized as “Ota Ward”.
　そして、仕分部２５０は、仕分けテーブルを参照することにより、搬送されて来た発送物を住所認識部２４０において認識した宛先住所に対応する区分箱に仕分けする。すなわち、「太田区」と誤植されている発送物は、「大田区」に対応する区分箱に適切に仕分けされる。 Then, the sorting unit 250 sorts the shipped items that have been conveyed into a sorting box corresponding to the destination address recognized by the address recognition unit 240 by referring to the sorting table. In other words, a shipment that is mistyped as “Ota Ward” is appropriately sorted into a sorting box corresponding to “Ota Ward”.
　以上のように、本実施形態に係る発送物区分機２００は、発送物の撮像画像から１文字ずつの文字画像データを取得し、各文字画像データの特徴量を演算して文字要素を決定することによりＩＤ列を生成する。１文字ずつの文字画像データごとに認識処理を行う場合、処理の負荷が小さいと共に高い精度でＩＤ列を生成することができる。また、発送物区分機２００は、生成したＩＤ列に対応する宛先住所を変換テーブルから抽出して出力する。この場合、出力不可となる可能性を低減できると共に誤った宛先住所が出力される可能性を低減できる。 As described above, the shipment sorting apparatus 200 according to the present embodiment obtains character image data for each character from the captured image of the shipment, calculates the feature amount of each character image data, and determines the character element. As a result, an ID string is generated. When the recognition process is performed for each character image data for each character, the processing load is small and the ID string can be generated with high accuracy. Further, the shipment sorting machine 200 extracts the destination address corresponding to the generated ID string from the conversion table and outputs it. In this case, it is possible to reduce the possibility that output is impossible, and it is possible to reduce the possibility that an incorrect destination address is output.
　従って、本実施形態に係る発送物区分機２００は、出力不可となる確率が低いと共に高い精度で宛先住所を出力して、発送物を最適な区分箱へ仕分けすることができる。 Therefore, the shipment sorting apparatus 200 according to the present embodiment has a low probability of being unable to output and outputs a destination address with high accuracy, and can sort the shipment into an optimum sorting box.
　また、分類テーブルには２３個の区名で使用されている３９文字およびこの３９文字に関する類似文字のみを登録すれば良い。分類テーブルに登録する文字数または単語数が増加すると、辞書作成の作業量が増加する他、類似文字も増加し、それらを区別する作業も増加する。「区」については２３区全区、「田」については、「千代田区」、「墨田区」、「世田谷区」に共通である。すなわち、「区」、「田」の辞書は複数の単語で共通に使用することができる。 In the classification table, only 39 characters used in 23 ward names and similar characters related to these 39 characters may be registered. When the number of characters or words registered in the classification table increases, the amount of work for creating a dictionary increases, the number of similar characters also increases, and the work for distinguishing them also increases. “Ward” is common to all 23 Wards, and “Ta” is common to “Chiyoda Ward”, “Sumida Ward”, and “Setagaya Ward”. That is, the “ku” and “ta” dictionaries can be used in common for a plurality of words.
　また、「大」、「田」、「区」等のように１文字毎に辞書を作成する場合、「大田区」等の単語毎に辞書を作成する場合と比較して、トータル的に、辞書作成に必要なサンプル数を多く確保することができる。 In addition, when creating a dictionary for each character such as “Ota”, “Ta”, “Ku”, etc., compared to creating a dictionary for each word such as “Ota Ward”, A large number of samples required for dictionary creation can be secured.
　ここで、上述の実施形態では、住所、番地および区分箱番号が関連付けられた仕分けテーブルを参照することにより、仕分部２５０が住所認識部２４０から出力された宛先住所を用いて発送物を仕分けしたが、これに限定されない。例えば、仕分部２５０がＩＤ列と区分箱番号とが関連付けられた仕分けテーブルを備えることもできる。この場合、住所認識部２４０は、生成したＩＤ列に対応する宛先住所を決定することなく、生成したＩＤ列をそのまま仕分部２５０に出力する。一方、仕分部２５０は、住所認識部２４０から出力されたＩＤ列を用いて発送物を仕分けする。 Here, in the above-described embodiment, the sorting unit 250 sorts the shipment using the destination address output from the address recognition unit 240 by referring to the sorting table in which the address, the address, and the sorting box number are associated. However, it is not limited to this. For example, the sorting unit 250 can include a sorting table in which an ID string and a sorting box number are associated with each other. In this case, the address recognition unit 240 outputs the generated ID string as it is to the sorting unit 250 without determining a destination address corresponding to the generated ID string. On the other hand, the sorting unit 250 sorts the shipment using the ID string output from the address recognition unit 240.
　上述した文字認識方法は、上述した文字認識の各工程をプログラム形式で記載して記録媒体等に記憶させ、このプログラムを文字認識装置１００および発送物区分機２００のＣＰＵが記録媒体から読み出すことによって実行することができる。 In the character recognition method described above, each step of character recognition described above is described in a program format and stored in a recording medium or the like, and this program is read out from the recording medium by the CPU of the character recognition device 100 and the shipment sorter 200. Can be executed.
　本願発明は上記実施形態に限定されるものではなく、この発明の要旨を逸脱しない範囲の設計の変更等があってもこの発明に含まれる。 The invention of the present application is not limited to the above-described embodiment, and any design change or the like within a range not departing from the gist of the invention is included in the invention.
　この出願は、２０１２年１月１９日に出願された日本出願特願２０１２－００８７９５を基礎とする優先権を主張し、その開示の全てをここに取り込む。 This application claims priority based on Japanese Patent Application No. 2012-008795 filed on January 19, 2012, the entire disclosure of which is incorporated herein.
　帳票からの単語読み取り作業や、発送物の仕分け作業等に利用することができる。 It can be used for reading words from forms and sorting out shipments.
　１０　　区分装置
　２０　　搬送装置
　３０　　文字認識装置
　３１　　単語テーブル
　３２　　読取手段
　３３　　区分手段
　３４　　解析手段
　３５　　認識手段
　４０　　仕分装置
　５０　　発送物
　１００　　文字認識装置
　１１０　　入力モジュール
　１２０　　文字切り出しモジュール
　１３０　　特徴抽出モジュール
　１４０　　文字分類モジュール
　１５０　　分類テーブル
　１６０　　変換モジュール
　１７０　　変換テーブル
　１８０　　出力モジュール
　１９０　　制御モジュール
　２００　　発送物区分機
　２１０　　供給部
　２２０　　搬送部
　２３０　　スキャナ部
　２４０　　住所認識部
　２５０　　仕分部
　２６０　　制御部 DESCRIPTION OF SYMBOLS 10 Classification apparatus 20 Conveyance apparatus 30 Character recognition apparatus 31 Word table 32 Reading means 33 Classification means 34 Analysis means 35 Recognition means 40 Sorting apparatus 50 Shipment thing 100 Character recognition apparatus 110 Input module 120 Character extraction module 130 Feature extraction module 140 Character classification module DESCRIPTION OF SYMBOLS 150 Classification table 160 Conversion module 170 Conversion table 180 Output module 190 Control module 200 Shipment sorter 210 Supply part 220 Conveyance part 230 Scanner part 240 Address recognition part 250 Sorting part 260 Control part

Claims (10)

  1. 複数の単語が所定の文字列と対応づけられて登録されている単語テーブルと、
    認識対象が表示されている表示体の画像情報を読み取る読取手段と、
    前記読み取った画像情報を単位画像に区分する区分手段と、
    前記区分した単位画像をそれぞれ解析して区分順に並べ、解析結果として出力する解析手段と、
    前記解析結果に対応する文字列を前記単語テーブルから抽出し、抽出した文字列に対応付けられている単語を前記認識対象の認識結果として出力する認識手段と、
    を備える文字認識装置。     A word table in which a plurality of words are registered in association with a predetermined character string;
    Reading means for reading image information of a display body on which a recognition target is displayed;
    A classifying means for classifying the read image information into unit images;
    Analyzing means for analyzing the segmented unit images and arranging them in the order of segmentation, and outputting them as analysis results;
    A recognition unit that extracts a character string corresponding to the analysis result from the word table, and outputs a word associated with the extracted character string as a recognition result of the recognition target;
    A character recognition device comprising:
  2. 類似の特徴量を有する文字同士がグループ化され、識別番号と対応付けられて登録されている文字テーブルをさらに備え、
    前記解析手段は、前記区分した単位画像の特徴量を取得し、取得した特徴量を有する文字が属するグループの識別番号を前記文字テーブルから抽出し、区分順に並べた識別番号を前記解析結果として出力する、
    請求項１記載の文字認識装置。     Characters having similar feature amounts are grouped together and further provided with a character table registered in association with an identification number,
    The analysis unit acquires a feature amount of the divided unit image, extracts an identification number of a group to which a character having the acquired feature amount belongs from the character table, and outputs an identification number arranged in the division order as the analysis result To
    The character recognition device according to claim 1.
  3. 前記文字テーブルには、類似の特徴量を有する文字の代わりに、グループの代表文字の特徴量が識別番号と対応付けられて登録されている、
    請求項２記載の文字認識装置。     In the character table, instead of characters having similar feature amounts, the feature amounts of group representative characters are registered in association with identification numbers.
    The character recognition device according to claim 2.
  4. 前記単語テーブルに類似の特徴量を有する文字同士を置き換えた単語がそれぞれ登録されている場合、前記類似の特徴量を有する文字同士は前記文字テーブルの異なるグループに登録される、請求項３記載の文字認識装置。 The words having the similar feature values are registered in different groups of the character table when words having the similar feature values replaced are registered in the word table, respectively. Character recognition device.
  5. 前記解析手段は、前記区分順に出力された解析結果に対応する文字列が前記単語テーブルに登録されていない場合、前記区分順に出力された解析結果に最も近い文字列を前記単語テーブルから抽出する、請求項１乃至４のいずれか１項記載の文字認識装置。 When the character string corresponding to the analysis result output in the classification order is not registered in the word table, the analysis unit extracts the character string closest to the analysis result output in the classification order from the word table. The character recognition device according to claim 1.
  6. 前記解析手段は、編集距離を用いることにより、前記区分順に出力された解析結果に最も近い文字列を前記単語テーブルから抽出する、請求項５記載の文字認識装置。 The character recognition device according to claim 5, wherein the analysis unit extracts, from the word table, a character string closest to the analysis result output in the order of division by using an edit distance.
  7. 前記区分手段は、前記読み取った画像情報から前記認識対象が表示されている領域を抜き出して、前記抜き出した領域の画像情報を単位画像に区分する、請求項１乃至６のいずれか１項記載の文字認識装置。 The said classification | category means extracts the area | region where the said recognition target is displayed from the read said image information, and classifies the image information of the extracted area | region into unit images. Character recognition device.
  8. 認識対象が表示されている発送物を搬送する搬送装置と、
    搬送中の前記発送物から画像情報を読み取り、前記認識対象の認識結果を出力する、請求項１乃至７のいずれか１項記載の文字認識装置と、
    前記出力された認識結果に基づいて、前記搬送されて来た発送物を仕分けする仕分装置と、
    を備える区分装置。     A transport device for transporting a shipment on which a recognition target is displayed;
    The character recognition device according to any one of claims 1 to 7, wherein image information is read from the shipment being conveyed, and a recognition result of the recognition target is output.
    Based on the output recognition result, a sorting device that sorts the shipped items that have been transported,
    A sorting apparatus comprising:
  9. 複数の単語が所定の文字列と対応づけられて登録されている単語テーブルを用いた文字認識であって、
    認識対象が表示されている表示体の画像情報を読み取り、
    前記読み取った画像情報を単位画像に区分し、
    前記区分した単位画像をそれぞれ解析して区分順に並べ、解析結果として出力し、
    前記解析結果に基づいて前記単語テーブルから文字列を抽出し、抽出した文字列に対応付けられている単語を前記認識対象の認識結果として出力する、
    文字認識方法。     Character recognition using a word table in which a plurality of words are registered in association with a predetermined character string,
    Read the image information of the display object on which the recognition target is displayed,
    The read image information is divided into unit images,
    Analyzing the segmented unit images and arranging them in the order of segmentation, outputting them as analysis results,
    A character string is extracted from the word table based on the analysis result, and a word associated with the extracted character string is output as a recognition result of the recognition target.
    Character recognition method.
  10. 複数の単語が所定の文字列と対応づけられて登録されている単語テーブルを備えた文字認識装置のコンピュータが実行する制御プログラムであって、
    認識対象が表示されている表示体の画像情報を読み取る機能、
    前記読み取った画像情報を単位画像に区分する機能、
    前記区分した単位画像をそれぞれ解析して区分順に並べ、解析結果として出力する機能、
    前記解析結果に基づいて前記単語テーブルから文字列を抽出し、抽出した文字列に対応付けられている単語を前記認識対象の認識結果として出力する機能、
    を前記コンピュータに実行させるための制御プログラム。     A control program executed by a computer of a character recognition device having a word table in which a plurality of words are registered in association with a predetermined character string,
    A function for reading image information of a display object on which a recognition target is displayed,
    A function of dividing the read image information into unit images;
    A function of analyzing the divided unit images and arranging them in the order of division, and outputting them as analysis results;
    A function of extracting a character string from the word table based on the analysis result and outputting a word associated with the extracted character string as a recognition result of the recognition target;
    A control program for causing the computer to execute.
PCT/JP2012/008210 2012-01-19 2012-12-21 Character recognition device, classifying device provided with same, character recognition method and control program WO2013108347A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2012-008795 2012-01-19
JP2012008795 2012-01-19

Publications (1)

Publication Number Publication Date
WO2013108347A1 true WO2013108347A1 (en) 2013-07-25

Family

ID=48798799

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2012/008210 WO2013108347A1 (en) 2012-01-19 2012-12-21 Character recognition device, classifying device provided with same, character recognition method and control program

Country Status (1)

Country Link
WO (1) WO2013108347A1 (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0676120A (en) * 1992-08-24 1994-03-18 Oki Electric Ind Co Ltd After-processing method for character recognizing result
JPH08180142A (en) * 1994-12-21 1996-07-12 Nec Corp Address reader
JPH08243505A (en) * 1995-03-14 1996-09-24 Hitachi Ltd Address reading device and its method
JPH09218921A (en) * 1996-02-13 1997-08-19 Oki Electric Ind Co Ltd General document reader
JP2006163830A (en) * 2004-12-07 2006-06-22 Nec Corp Character recognizing device, character recognizing method, and character recognizing program
JP2007052700A (en) * 2005-08-19 2007-03-01 Toshiba Corp Machine translation apparatus and machine translation program
JP2007122403A (en) * 2005-10-28 2007-05-17 Fuji Xerox Co Ltd Device, method, and program for automatically extracting document title and relevant information
WO2011036830A1 (en) * 2009-09-24 2011-03-31 日本電気株式会社 Word recognition device, method, non-transitory computer readable medium storing program and shipped item classification device

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0676120A (en) * 1992-08-24 1994-03-18 Oki Electric Ind Co Ltd After-processing method for character recognizing result
JPH08180142A (en) * 1994-12-21 1996-07-12 Nec Corp Address reader
JPH08243505A (en) * 1995-03-14 1996-09-24 Hitachi Ltd Address reading device and its method
JPH09218921A (en) * 1996-02-13 1997-08-19 Oki Electric Ind Co Ltd General document reader
JP2006163830A (en) * 2004-12-07 2006-06-22 Nec Corp Character recognizing device, character recognizing method, and character recognizing program
JP2007052700A (en) * 2005-08-19 2007-03-01 Toshiba Corp Machine translation apparatus and machine translation program
JP2007122403A (en) * 2005-10-28 2007-05-17 Fuji Xerox Co Ltd Device, method, and program for automatically extracting document title and relevant information
WO2011036830A1 (en) * 2009-09-24 2011-03-31 日本電気株式会社 Word recognition device, method, non-transitory computer readable medium storing program and shipped item classification device

Similar Documents

Publication Publication Date Title
US11348353B2 (en) Document spatial layout feature extraction to simplify template classification
KR100324847B1 (en) Address reader and mails separater, and character string recognition method
US8045798B2 (en) Features generation and spotting methods and systems using same
EP1362322B1 (en) Holistic-analytical recognition of handwritten text
EP3001352B1 (en) Image processing apparatus and image processing method
US5787194A (en) System and method for image processing using segmentation of images and classification and merging of image segments using a cost function
JP3485020B2 (en) Character recognition method and apparatus, and storage medium
JP2012500428A (en) Segment print pages into articles
JP2015058390A (en) Sorting device and sorting method
JP4855698B2 (en) Address recognition device
Barnouti et al. An efficient character recognition technique using K-nearest neighbor classifier
WO2013108347A1 (en) Character recognition device, classifying device provided with same, character recognition method and control program
JP2012252657A (en) Business form identification system, business form identification method and business form processing program
FI3903230T3 (en) Structural image matching by hashing descriptors of singularities of the gradient
Kumar et al. Line based robust script identification for indianlanguages
Duth et al. Recognition of hand written and printed text of cursive writing utilizing optical character recognition
JP5178851B2 (en) Address recognition device
Afifah et al. The line segmentation algorithm of Indonesian electronic identity card (e-KTP) for data digitization
Halder et al. Individuality of Bangla numerals
Mazzei et al. Extraction and classification of handwritten annotations
Sarkar et al. Classifying foreground pixels in document images
Jain et al. A NOVEL CUTTING EDGE NATURE INSPIRED OPTIMIZATION OF BRAHMI CHARACTERS RECOGNITION
Nigam et al. Novel System for Retrieval of Composite Trademarks Using Multi-feature Voting
Andreas Optical character recognition (OCR) on electrical specification plates
JPH0793474A (en) Character area detector

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 12866208

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 12866208

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP