JP2016167166A - Destination recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a destination recognition device which can specify a destination region with higher accuracy from among a plurality of destination region candidates when performing destination recognition of a delivery object.SOLUTION: A destination region device comprises: an image processing part which generates image data of a delivery object surface; a number keyword input reception part which receives input of a number keyword being a number allocated for each address zone in accordance with the image data; a destination region candidate detection processing part which detects a destination region candidate being a region including a character string from the image data; a destination region pattern learning processing part which detects an address keyword being a term constituting the number keyword and address from the destination region candidate and generates information indicating the physical relationship between the number keyword and the address keyword; and a destination region pattern database which registers information indicating the physical relationship between the number keyword and the address keyword as a destination region pattern.SELECTED DRAWING: Figure 1

Description

  Embodiments described herein relate generally to a destination recognition apparatus that recognizes a destination of a delivery object.

  A conventional destination recognition device uses optical character recognition (OCR) to acquire destination information by reading a character string from an image of a delivery object of a postal letter or a parcel and converting it into a text. However, when the destination of the delivery object is recognized, both the sender area information and the destination area information are candidates for the destination area, and therefore it may not be possible to identify which area of the candidates is the destination area. is there.

  For this reason, a technique for specifying the destination area with higher accuracy is desired.

JP 2007-4584 A

  Therefore, the problem to be solved by the present invention is to provide a destination recognition device that can specify a destination area with higher accuracy from a plurality of destination area candidates when performing destination recognition of a delivery object. .

  In order to solve the above problems, an address recognition apparatus according to an embodiment reads an image of a delivery object surface and generates image data of the delivery object, and image data generated by the image processing part A number keyword input receiving unit for inputting a number keyword that is a number assigned to each address area, a number keyword / corresponding address dictionary unit for storing a corresponding address corresponding to the number keyword, and the image data From the destination area candidate detection processing unit that detects a destination area candidate that is an area including a character string, and the number received by the number keyword input reception unit from the destination area candidates detected by the destination area candidate detection processing unit A keyword is detected, and the number keyword input acceptance among the number keywords stored in the number keyword / corresponding address dictionary unit Acquires a corresponding address corresponding to the number keyword received by the address, detects an address keyword that is a term constituting the address from the acquired corresponding address, and a positional relationship between the detected number keyword and the detected address keyword A destination area pattern learning processing unit for generating information indicating the information, and a destination area pattern database in which information indicating the positional relationship between the number keyword and the address keyword generated by the destination area pattern learning processing unit is registered as a destination area pattern; Are provided.

  The destination recognition apparatus of the embodiment detects a number keyword and an address keyword from destination area candidates detected by the destination area candidate detection processing unit, and displays information indicating a positional relationship between the detected number keyword and address keyword. The destination area detected by the destination area candidate detection processing unit is generated by collating the information indicating the positional relationship between the generated number keyword and address keyword with a registered destination area pattern registered in the destination area pattern database. Information indicating the positional relationship between the generated number keyword and address keyword by collation by the destination address determination processing unit that selects a destination area candidate character string as a destination address from among all candidates, and the destination address determination processing unit If there is a registered destination area pattern that matches Information indicating a positional relationship between the number keyword and address keywords turn comprises a collating result storage unit stored as the collation result.

It is a functional block diagram of the destination recognition apparatus 100 which concerns on embodiment of this invention. It is a figure which shows an example of the flow of a destination area | region pattern learning process. It is a figure which shows an example of the flow of a destination address determination process. 6 is a diagram illustrating an example of data registered in a destination area pattern database 33. FIG. 4 is an example of data stored in a matching result storage unit 34. It is a figure which shows an example of the flow of a destination area | region candidate narrowing-down process. 7 is an example of data stored in a destination area candidate narrowing result storage unit 35. It is a figure which shows the example which detects the destination area | region candidate C from the image data P of a delivery target object. It is a figure which shows the example which produces | generates the information (positional relationship information R) which shows the positional relationship of a number keyword and an address keyword from the destination area | region candidate C from which the number keyword and the address keyword were detected.

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings.

  With reference to FIG. 1, the configuration of a destination recognition apparatus 100 according to an embodiment of the present invention will be described. FIG. 1 is a functional block diagram illustrating a configuration example of a destination recognition apparatus 100 according to an embodiment of the present invention.

  The destination recognition apparatus 100 according to the embodiment mainly includes a functional unit group used for destination area pattern learning processing that is executed before the destination recognition apparatus 100 is operated, and a destination address determination that is performed when the destination recognition apparatus 100 is operated. And a functional unit group used for processing.

  The functional unit group used for the destination area pattern learning process performs a process for accumulating data necessary for operation before the operation of the destination recognition apparatus 100 used for the delivery object sorter or the like. The destination area pattern learning process includes an image reading unit 11, an image storage unit 31, an image reading control unit 21, a destination area candidate detection processing unit 22, a number keyword input receiving unit 12, a number keyword / corresponding address dictionary unit 32, a destination area. A pattern learning processing unit 23 and a destination area pattern database 33 are used.

  The image reading unit 11 captures an image of the surface of a delivery object such as a postal letter or a parcel. The image reading unit 11 is, for example, a scanner or a camera, and captures the surface of the delivery object one by one, and generates image data P that is data of the imaged surface of the delivery object. The image reading unit 11 is controlled by the image reading control unit 21.

  The image storage unit 31 stores the image data P generated by the image reading unit 11 by the image reading control unit 21. The image data P stored in the image storage unit 31 is referred to by a destination area candidate detection processing unit 22 described later. The image data P stored in the image storage unit 31 is referred to by the number keyword input receiving unit 12 described later.

  The image reading control unit 21 controls the operations of the image reading unit 11 and the image storage unit 31.

  The destination region candidate detection processing unit 22 refers to the image data P stored in the image storage unit 31 and is a region including a character string that is a candidate for a region in which a destination is described from the referenced image data P. The destination area candidate C is detected. Further, the destination area candidate detection processing unit 22 outputs information on the detected destination area candidate C to the destination area pattern learning processing unit 23 described later.

  The number keyword input receiving unit 12 is a number keyword that is a number assigned to each address area, such as a Japanese “zip code” or an American “ZIP code”, for each image data P stored in the image storage unit 31. The input is accepted.

  The operator refers to the image data P stored in the image storage unit 31, and uses the user interface such as a keyboard to number the number keyword included in the destination area candidate C determined as the destination area from the referenced image data P. Input to the keyword input receiving unit 12.

  In addition, the number keyword input receiving unit 12 outputs information on the input number keyword to the destination area pattern learning processing unit 23 described later.

  The number keyword / corresponding address dictionary unit 32 stores a number keyword and a corresponding address corresponding to the number keyword in advance. For example, in Japan, a corresponding address from “zip code” to “XX city XX town” can be specified. In the case of the United States, the corresponding address from “ZIP code” to “XX City” can be specified.

  The number keyword / corresponding address dictionary unit 32 stores the correspondence between these number keywords and corresponding addresses. The number keyword / corresponding address dictionary unit 32 is referred to by the destination area pattern learning processing unit 23 described later, and the information of the number keyword output to the destination region pattern learning processing unit 23 described later is collated with the corresponding corresponding address. The

  The destination area pattern learning processing unit 23 mainly generates positional relationship information R described later, and registers a destination area pattern A described later in a destination area pattern database 33 described later.

  First, the destination area pattern learning processing unit 23 detects the number keyword input by the operator to the number keyword input receiving unit 12 from each destination area candidate C detected by the destination area candidate detection processing unit 22. This is because the destination area candidate C having the number keyword input by the operator to the number keyword input receiving unit 12 is most likely to correspond to the destination area.

  The destination area pattern learning processing unit 23 further refers to the number keyword / corresponding address dictionary unit 32 for the destination area candidate C having the number keyword, and the number keyword input by the operator to the number keyword input receiving unit 12 and the number The number keyword and the corresponding address stored in the keyword / corresponding address dictionary unit 32 are collated, and the corresponding address is acquired.

  The destination area pattern learning processing unit 23 further selects “city”, “town”, “street”, “building”, “post office box” from the corresponding addresses of the number keyword acquired from the number keyword / corresponding address dictionary unit 32. The address keyword, which is a term constituting the address, is detected.

  The destination area pattern learning processing unit 23 further selects a character string such as “To”, “Sama”, “Mr.”, “Mr”, and the like from the destination area candidate C having the number keyword and the address keyword of the corresponding address. A specific keyword that is a term that suggests a certain keyword is detected. The destination area candidate C may or may not include a specific keyword.

  Then, the destination area pattern learning processing unit 23 generates information indicating the positional relationship between the number keyword and the address keyword from the destination area candidate C having the number keyword and the address keyword of the corresponding address. This is because the destination area candidate C having the number keyword and the address keyword of the corresponding address input to the number keyword input receiving unit 12 by the operator is most likely to correspond to the destination area among a plurality of detected destination area candidates C. is there. If the specific keyword is included in the destination area candidate C, the destination area pattern learning processing unit 23 generates information indicating the positional relationship among the specific keyword, the number keyword, and the address keyword. The information indicating the positional relationship between the number keyword and the address keyword and the information indicating the positional relationship between the specific keyword, the number keyword, and the address keyword are collectively referred to as positional relationship information R hereinafter.

  The positional relationship information R indicates information indicating the mutual positional relationship between keywords included in the destination area candidate C. As an example, when a specific keyword is included in the destination area candidate, the positional relationship information R indicates a relative positional relationship representing the direction, distance, and the like from the number keyword to the specific keyword and each address keyword. In addition to this, a specific keyword, a number keyword, and a combination of position information representing the coordinates of each address keyword may be used.

  Then, the destination area pattern learning processing unit 23 registers or updates the type of the generated positional relationship information R as the destination area pattern A in the destination area pattern database 33 described later.

  In the destination area pattern database 33, the destination area pattern A that is the type of the positional relationship information R is, for example, a destination area pattern number that is a number for identifying the destination area pattern A, and the destination area pattern A is a destination area pattern learning process. It is registered or updated by the destination area pattern learning processing unit 23 together with a score value indicating whether or not the number of appearances is large within a predetermined period. The destination area pattern database 33 may not include a score value.

  The destination area pattern A is a type of information indicating the positional relationship between the number keyword and the address keyword, or a type of information indicating the positional relationship between the specific keyword, the number keyword, and the address keyword.

  The score value is a value that is updated as needed depending on how many times the destination area pattern A appears within a predetermined period when the destination area pattern is learned. Here, the predetermined period may be an entire period from when the destination recognition apparatus starts to be used, or a specific period. Also, here, the destination area pattern A having a higher number of appearances has a higher score value.

  For this reason, it can be said that the more likely destination area pattern A has a higher likelihood that it is the destination area. That is, when a plurality of pieces of positional relationship information R that matches a plurality of destination area patterns A is generated from a plurality of destination area candidates C in a destination address determination process described later, a score among the plurality of destination area candidates C is scored. The destination area candidate C that has generated the positional relationship information R that matches the destination area pattern A having the highest value has the highest likelihood of being the destination area. For this reason, the character string of the destination area candidate C in which the destination area pattern A having the highest score value is detected is output as the destination address.

  The destination area pattern database 33 is referred to by the destination area pattern learning processing unit 23. That is, the type of the positional relationship information R generated by the destination region pattern learning processing unit 23 is collated with the destination region pattern A registered in the destination region pattern database 33 that matches the type of the positional relationship information R. .

  When the type of the positional relationship information R generated by the destination region pattern learning processing unit 23 is not registered in the destination region pattern database 33, the type of the unregistered positional relationship information R is the new destination region pattern A. Is registered in the destination area pattern database 33 by the destination area pattern learning processing unit 23.

  On the other hand, when the type of the positional relationship information R generated by the destination area pattern learning processing unit 23 has already been registered in the destination area pattern database 33, a predetermined numerical value is added to the score value of the registered destination area pattern A. By adding, the contents of the data in the destination area pattern database 33 are updated by the destination area pattern learning processing unit 23.

  The destination area pattern A data is accumulated in the destination area pattern database 33 by the destination area pattern learning process by the functional unit group described above.

  The functional unit group used for the destination address determination process is a delivery object sorting machine or the like after the data of the destination area pattern A is accumulated in the destination area pattern database 33 by the functional unit group used for the destination area pattern learning process described above. Destination address determination processing during operation of the destination recognition apparatus 100 used in the above-described process. This destination address determination processing includes an image reading unit 11, an image storage unit 31, an image reading control unit 21, a destination region candidate detection processing unit 22, a number keyword / corresponding address dictionary unit 32, a destination address determination processing unit 24, a destination region pattern. A database 33, a matching result storage unit 34, a destination area candidate narrowing result storage unit 35, and a destination address output unit 13 are used. The destination area candidate narrowing result storage unit 35 may be omitted or not used.

  Here, the image reading unit 11, the image storage unit 31, the image reading control unit 21, the destination area candidate detection processing unit 22, the number keyword / corresponding address dictionary unit 32, and the destination area pattern database 33 are the destinations. It is used in common with the functional unit group used for the area pattern learning process and the functional unit group used for the destination address determination process. Among these, the image reading unit 11, the image storage unit 31, and the image reading control unit 21 have the same configuration and function when the destination area pattern learning process is executed and when the destination address determination process is executed. Therefore, the description is omitted here.

  The destination area candidate detection processing unit 22 refers to the image data P stored in the image storage unit 31 and detects a destination area candidate C from the referenced image data P. The destination area candidate detection processing unit 22 outputs information of the detected destination area candidate C to a destination address determination processing unit 24 described later.

  The number keyword / corresponding address dictionary unit 32 is referred to by a destination address determination processing unit 24 to be described later, and the information of the number keyword detected by the destination address determination processing unit 24 to be described later is collated with the corresponding corresponding address.

  The destination address determination processing unit 24 mainly generates the positional relationship information R, selects a character string serving as the destination address from the character string of the destination area candidate C, and outputs the destination address to the destination address output unit 13 described later. To do. Further, the destination address determination processing unit 24 is a processing entity when executing a destination area candidate narrowing-down process described later.

  First, the destination address determination processing unit 24 detects a number keyword and an address keyword from each destination region candidate C detected by the destination region candidate detection processing unit 22. That is, first, the destination address determination processing unit 24 detects a number keyword. The destination address determination processing unit 24 further refers to the number keyword / corresponding address dictionary unit 32 for the destination area candidate C having the number keyword, and stores the detected number keyword and the number keyword / corresponding address dictionary unit 32 in the number keyword / corresponding address dictionary unit 32. The corresponding number keyword and the corresponding address are collated to obtain the corresponding address. The destination area pattern learning processing unit 23 further detects an address keyword among the corresponding addresses of the number keyword acquired from the reference of the number keyword / corresponding address dictionary unit 32.

  The destination address determination processing unit 24 further detects a specific keyword from the destination area candidate C having the number keyword and the address keyword of the corresponding address. The specific keyword may or may not be included in the destination area candidate C.

  Then, the destination address determination processing unit 24 generates the positional relationship information R from all the destination area candidates C having the number keyword and the address keyword of the corresponding address among the respective destination area candidates C.

  The destination address determination processing unit 24 sequentially collates the type of the generated positional relationship information R with the destination area pattern A registered in the destination area pattern database 33 and its score value. The destination area pattern database 33 may not include a score value.

  The destination address determination processing unit 24 stores the verification result in the verification result storage unit 34 described later. That is, when the destination area pattern database 33 has a destination area pattern A that matches the type of the positional relationship information R generated by the destination address determination processing unit 24, for example, The destination area pattern number, the score value of the destination area pattern, and the ranking information ranked in descending order of the score value are stored.

  When the destination address determination processing unit 24 finishes generating the positional relationship information R for all the destination region candidates C, the destination address determination processing unit 24 refers to a matching result storage unit 34 described later, for example, the destination region pattern A having the highest score value. The character string of the destination area candidate C that generated the matching positional relationship information R is selected, and this is determined as the destination address. The destination address determination processing unit 24 outputs this destination address to the destination address output unit 13 described later.

  In the destination area pattern database 33, the destination area pattern A is registered together with the score value by the destination area pattern learning process described above. The destination area pattern database 33 is referred to by the destination address determination processing unit 24. That is, the type of the positional relationship information R generated by the destination address determination processing unit 24 is collated with the destination region pattern A registered in the destination region pattern database 33 that matches the type of the positional relationship information R.

  Note that the destination address determination processing unit 24, as in the case of executing a destination area candidate narrowing process described later, is a destination address such as “From”, “Return”, “Forward”, etc. In some cases, the NG keyword, which is a term that suggests other than the above, is detected. By performing the destination area candidate narrowing process prior to the destination address determination process, for example, it becomes possible to narrow down the destination area candidate C in advance. In this stage of performing the destination area candidate narrowing process, the destination area candidate to be processed is referred to as a primary destination area candidate C ′.

  In this case, the destination address determination processing unit 24 stores a destination area candidate narrowing result to be described later in a destination area candidate narrowing result storage unit 35 to be described later. The destination area candidate narrowing down result is that information regarding the presence or absence of the NG keyword is stored for all primary destination area candidates C ′.

  The collation result storage unit 34 stores the result of the destination address determination processing unit 24 referring to the destination area pattern database 33 by the destination address determination processing unit 24. That is, when the destination area pattern A that matches the type of the positional relationship information R generated by the destination address determination processing unit 24 exists in the destination area pattern database 33, for example, the destination area pattern The number, the score value of the destination area pattern, and the ranking information ranked in descending order of the score value are stored. These collation results are referred to by the destination address determination processing unit 24.

  The destination area candidate narrowing result storage unit 35 performs a destination address candidate narrowing process, which will be described later, for each primary destination area candidate C ′ detected by the destination area candidate detection processing unit 22. The result of the determination processing unit 24 detecting the NG keyword, that is, the information on the presence / absence of the NG keyword is stored by the destination address determination processing unit 24.

  For example, the destination address output unit 13 has the highest score value among the destination area pattern A and the score value stored in the matching result storage unit 34, or the highest rank of the score value. The character string of the destination area candidate C for which the positional relationship information R that matches the destination area pattern A is generated is received from the destination address determination processing unit 24 as the destination address.

  The destination address output unit 13 may directly output the received destination address to the delivery object sorting machine side or may output it to a delivery database provided in the destination recognition apparatus 100.

  The destination recognition of the delivery object is executed by the destination address determination processing by the above functional unit group. The destination recognition apparatus 100 includes an interface unit 10, a control unit 20, and a storage unit 30. The interface unit 10 includes an image reading unit 11, a number keyword input receiving unit 12, and a destination address output unit 13. The control unit 20 includes an image reading control unit 21, a destination area candidate detection processing unit 22, a destination area pattern learning processing unit 23, and a destination address determination processing unit 24. The storage unit 30 includes an image storage unit 31, a number keyword / corresponding address dictionary unit 32, a destination area pattern database 33, a matching result storage unit 34, and a destination area candidate narrowing result storage unit 35.

  Here, the functional units of the image reading control unit 21, the destination area candidate detection processing unit 22, the destination area pattern learning processing unit 23, and the destination address determination processing unit 24 of the control unit 20 are, for example, a CPU (Central Processing Unit). Etc.) is a software function unit that functions by executing a program stored in the storage unit 30. Also, some or all of these functional units may include hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), and various interfaces.

  Here, the process of learning the destination area pattern A will be described with reference to FIG. FIG. 2 is a flowchart illustrating an example of the flow of the destination area pattern learning process. Before the start of the process of learning the destination area pattern A, the image acquisition unit as a functional unit group that executes the functions realized by the image reading unit 11, the image storage unit 31, and the image reading control unit 21. The image data P of the delivery object is acquired.

  The destination area candidate detection processing unit 22 analyzes the image data P stored in the image storage unit 31, and detects all the destination area candidates C (step S101). An arbitrary method is used as the detection method. For example, a method is used in which line candidates are extracted from the arrangement of character candidates, and the destination area candidate C is detected from the arrangement of line candidates. In addition, a method of connecting adjacent character candidates and detecting the connection result as a destination area candidate C may be used.

  Each destination area candidate C is assigned a serial number from 0 to n (n is a natural number). This is represented by a variable i set from the initial value 0. The destination area pattern learning processing unit 23 sequentially performs processing from the 0th destination area candidate C (step S102).

  The destination area pattern learning processing unit 23 determines whether or not the number keyword input in advance to the number keyword input receiving unit 12 by the operator is detected from the destination area candidate C to be processed (step S103).

  When the number keyword is not detected (NO in step S103), the destination area pattern learning processing unit 23 performs destination area pattern learning processing on all destination area candidates C detected by the destination area candidate detection processing unit 22. It is determined whether or not the process has been completed (step S112). If the processing for all destination area candidates C has not been completed (NO in step S112), the processing proceeds to the i + 1th destination area candidate C (step S113).

  On the other hand, when the number keyword is detected (step S103), the destination area pattern learning processing unit 23 refers to the number keyword / corresponding address dictionary unit 32 for the destination area candidate C having the number keyword, and the operator selects the number. The number keyword input to the keyword input reception unit 12 is compared with the number keyword and the corresponding address stored in the number keyword / corresponding address dictionary unit 32 to acquire the corresponding address (step S104).

  Subsequently, the destination area pattern learning processing unit 23 determines whether or not an address keyword of the corresponding address is detected from the destination area candidate C having the number keyword (step S105).

  When the address keyword is not detected (NO in step S105), the destination area pattern learning processing unit 23 performs the destination area pattern learning process for all destination area candidates C detected by the destination area candidate detection processing unit 22. It is determined whether or not the process has been completed (step S112). If the processing for all destination area candidates C has not been completed (NO in step S112), the process proceeds to the (i + 1) th destination area candidate C (step S113).

  On the other hand, when an address keyword is detected (YES in step S105), the destination area pattern learning processing unit 23 further detects a specific keyword from the destination area candidate C having the number keyword and the address keyword of the corresponding address. Is determined (step S106).

  When the specific keyword is not detected (NO in step S106), the destination area pattern learning processing unit 23 generates information indicating the positional relationship between the number keyword and the address keyword (step S107). On the other hand, when the specific keyword is detected (YES in step S106), the destination area pattern learning processing unit 23 generates information indicating the positional relationship among the specific keyword, the number keyword, and the address keyword (step S108). That is, in step S107 or step S108, the destination area pattern learning processing unit 23 generates the positional relationship information R.

  Subsequently, the destination area pattern learning processing unit 23 checks the type of the generated positional relationship information R with the destination area pattern A registered in the destination area pattern database 33, and determines whether there is a matching destination area pattern A. Is determined (step S109).

  As a result of the collation, when the type of the generated positional relationship information R does not match any of the registered destination region patterns A (NO in step S109), the destination region pattern learning processing unit 23 determines the type of the positional relationship information R. Are registered in the destination area pattern database 33 as a new destination area pattern A (step S110).

  On the other hand, when the type of the generated positional relationship information R matches the registered destination area pattern A (YES in step S109), the destination area pattern learning processing unit 23 calculates the score value of the registered destination area pattern A. A predetermined numerical value is added to. Thereby, the content of the data of the destination area pattern A is updated (step S111). An arbitrary method is used as the score value adding method.

  Subsequently, the destination area pattern learning processing unit 23 determines whether or not the destination area pattern learning process has been completed for all destination area candidates C detected by the destination area candidate detection processing unit 22 (step S112). ). If the processing for all destination area candidates C has not been completed (NO in step S112), the process proceeds to the (i + 1) th destination area candidate C (step S113).

  On the other hand, if the processing for all destination area candidates C has been completed (YES in step S112), the destination area pattern learning process is terminated.

  Thus, the destination area pattern learning process for the image data P of a certain delivery object acquired by the image acquisition unit is completed. By repeating this destination area pattern learning process, the data of the destination area pattern A is accumulated in the destination area pattern database 33.

  Next, processing for determining a destination address will be described with reference to FIG. FIG. 3 is a flowchart illustrating an example of the flow of a destination address determination process. In order to execute this processing, it is assumed that the destination area pattern learning process described above is executed for a certain amount of delivery objects, and the destination area pattern database 33 has accumulated a sufficient amount of data.

  Before the start of the process of determining the destination address, the image acquisition unit as a functional unit group that executes the functions realized by the image reading unit 11, the image storage unit 31, and the image reading control unit 21 Image data P of the object is acquired.

  The destination area candidate detection processing unit 22 analyzes the image data P stored in the image storage unit 31, and detects all the destination area candidates C (step S201).

  Each destination area candidate C is assigned a serial number from 0 to n (n is a natural number). This is represented by a variable i set from the initial value 0. The destination address determination processing unit 24 sequentially performs processing from the 0th destination area candidate C (step S202).

  The destination address determination processing unit 24 determines whether or not the number keyword and the address keyword of the corresponding address are detected from the processing target destination area candidate C (step S203). Specifically, first, the destination address determination processing unit 24 detects a number keyword from the destination area candidate C to be processed, and refers to the number keyword / corresponding address dictionary unit 32 to detect the detected number keyword. And the address keyword of the corresponding address is detected from the destination area candidate C to be processed.

  When the number keyword and the address keyword of the corresponding address are not detected (NO in step S203), the destination address determination processing unit 24 sends destinations to all destination region candidates C detected by the destination region candidate detection processing unit 22. It is determined whether or not the address determination process is completed (step S209). If the processing for all destination area candidates C has not been completed (NO in step S209), the process proceeds to the (i + 1) th destination area candidate C (step S211).

  On the other hand, when the number keyword and the address keyword are detected (YES in step S203), the destination address determination processing unit 24 further detects a specific keyword from the destination area candidate C having the address keyword of the number keyword and the corresponding address. It is determined whether or not to be performed (step S204).

  When the specific keyword is not detected (NO in step S204), the destination address determination processing unit 24 generates information indicating the positional relationship between the number keyword and the address keyword (step S205). On the other hand, when the specific keyword is detected (YES in step S204), the destination address determination processing unit 24 generates information indicating the positional relationship among the specific keyword, the number keyword, and the address keyword (step S206). That is, in step S205 or step S206, the destination area pattern learning processing unit 23 generates the positional relationship information R.

  Subsequently, the destination address determination processing unit 24 checks the type of the generated positional relationship information R with the destination area pattern A registered in the destination area pattern database 33, and determines whether there is a matching destination area pattern A. Is determined (step S207).

  As a result of the collation, when the type of the generated positional relationship information R does not match any of the registered destination area patterns A (NO in step S207), the destination address determination processing unit 24 uses the destination area candidate detection processing unit 22 to It is determined whether or not the destination address determination process for all the detected destination area candidates C has been completed (step S209). If the processing for all destination area candidates C has not been completed (NO in step S209), the processing proceeds to the i + 1-th destination area candidate C (step S211).

  On the other hand, when there is a destination area pattern A that matches the type of the generated positional relationship information R (YES in step S207), the destination address determination processing unit 24 stores the matching result in the matching result storage unit 34 ( Step S208). Specifically, for example, for the matched destination area pattern A, the destination area pattern number and the score value are stored in the matching result storage unit 34 together with the ranking of the score values. Note that the ranking of score values may not be stored.

  Subsequently, the destination address determination processing unit 24 determines whether or not the destination address determination processing has been completed for all destination region candidates C detected by the destination region candidate detection processing unit 22 (step S209). If the processing for all destination area candidates C has not been completed (NO in step S209), the processing proceeds to the i + 1-th destination area candidate C (step S211).

  On the other hand, when the processing for all the destination area candidates C is completed (YES in step S209), the destination address determination processing unit 24 uses, for example, the score among the destination area patterns A stored in the matching result storage unit 34. The character string of the destination area candidate C from which the highest value, that is, the one with the highest rank is selected, is determined as the destination address, and is output to the destination address output unit 13 (step S210).

  Thus, the destination address determination process is completed. That is, the destination address is determined from the score value with the destination area pattern A learned before operation.

  Here, with reference to FIG. 4, data registered in the destination area pattern database 33 that is commonly used in the destination area pattern learning process and the destination address determination process will be described. FIG. 4 is an example of data registered in the destination area pattern database 33.

  In the destination area pattern database 33, for example, a destination area pattern A that is a type of the positional relationship information R is registered together with a destination area pattern number and a score value. The destination area pattern number is an identification number assigned to each destination area pattern A.

  In addition, it has described as a figure which shows the concept of the positional relationship information R here. Actually, the positional relationship information R may be expressed as a relative positional relationship that represents the direction and distance of each keyword that constitutes the destination area pattern A, or may be expressed by each positional information that represents the coordinates of each keyword. It may be expressed. The same applies to the positional relationship information R in FIG.

  In this example, the destination area pattern A with the destination area pattern number 002 has the highest score value. This means that it appears most frequently in the destination area pattern learning processing unit 23. For example, when the positional relationship information R that matches the destination region pattern number 002 is generated in the destination address determination process, the character string of the destination region candidate C that generated the positional relationship information R that matches the destination region pattern number 002 is It can be seen that the likelihood of being a destination address is high.

  Next, data stored in the matching result storage unit 34 used for the destination address determination process will be described with reference to FIG. FIG. 5 is an example of data stored in the matching result storage unit 34.

  The collation result storage unit 34 stores, for example, the destination area pattern numbers and score values of all the matching destination area patterns A as a result of the collation for each destination area candidate C for which the destination address determination processing unit 24 has generated the positional relationship information R. Are stored together with the ranking of the score values. Note that the ranking of score values may not be stored.

  In this example, among the five destination area candidates C from the destination area candidate No. 0 to the destination area candidate No. 4 (or more), the positional relationship information R that matches the registered destination area pattern A is generated. Are destination area candidate No. 0 and destination area candidate No. 4. When each score value is ranked, the destination area candidate No. 4 has a higher rank (score value is higher). Therefore, in this example, the destination area candidate No. 4 character string is selected, it is determined that this is the destination address, and is output to the destination address output unit 13 as the destination address (step S210 in FIG. 3).

  Next, a process for narrowing down the destination area candidates C in the destination address determination process will be described with reference to FIG. FIG. 6 is a diagram illustrating an example of the flow of destination area candidate narrowing processing. By performing this process prior to the destination address determination process, for example, it is possible to narrow down in advance the destination area candidates C to be processed in step S201 in FIG.

  First, the destination area candidate detection processing unit 22 analyzes the image data P stored in the image storage unit 31, and detects all the first destination area candidates C ′ (step S301).

  Each primary destination area candidate C ′ is assigned a serial number from 0 to n (n is a natural number). This is represented by a variable i set from the initial value 0. The destination address determination processing unit 24 sequentially performs processing from the 0th first destination area candidate C ′ (step S <b> 302).

  The destination address determination processing unit 24 determines whether or not the NG keyword is detected from the first destination area candidate C ′ to be processed (step S <b> 303).

  When the NG keyword is detected (YES in step S303), the destination address determination processing unit 24 narrows down the destination area candidate information on the serial number of the primary destination area candidate C ′ and the information that the NG keyword is detected. The result is stored in the result storage unit 35 (step S304).

  On the other hand, when the NG keyword is not detected, the destination address determination processing unit 24 determines whether or not the destination area candidate narrowing-down process has been completed for all the first destination area candidates C ′ (step S1). S305). If all the primary destination area candidates C ′ have not been processed (NO in step S305), the process proceeds to the (i + 1) th first destination area candidate C ′ (step S307).

  When the processing for all the primary destination area candidates C ′ has been completed (YES in step S305), the destination address determination processing unit 24 selects the first one having the NG keyword among the primary destination area candidates C ′. The next destination area candidate C ′ is determined as a non-destination area that does not correspond to the destination area, and is excluded from the destination area candidate C. In this way, the destination address determination processing unit 24 selects and selects the destination area candidate C from the primary destination area candidates C ′ (step S306).

  Thus, the destination area candidate narrowing process is completed. When used together with the above-described destination address determination process, the selected destination area candidate C is used as the destination area candidate C in step S201 of the destination address determination process, and the destination address determination process is executed.

  Since the destination area candidate narrowing-down process is performed prior to the destination address determination process, it is a non-destination area, so it is not necessary to execute the destination address determination process for the non-target destination area candidate C. Increases processing accuracy. Furthermore, since the number of destination area candidates C to be processed in the destination address determination process is reduced, the time until the destination address determination process is completed is shortened.

  Next, data stored in the destination area candidate narrowing result storage unit 35 used for the destination area candidate narrowing processing will be described with reference to FIG. FIG. 7 is an example of data stored in the destination area candidate narrowing result storage unit 35.

  The destination area candidate narrowing result storage unit 35 stores information on the presence / absence of an NG keyword for each primary destination area candidate C ′. In this example, as a result of the destination address determination processing unit 24 detecting NG keywords for the five primary destination area candidates C ′, those having no NG keyword are the first destination area candidate No. 0, They are the first destination area candidate No. 3 and the first destination area candidate No. 4. Therefore, in this example, the first destination area candidate No. 1 and the first destination area candidate No. 2 having the NG keyword are determined as non-destination areas. For this reason, only the first destination area candidate No. 0, the first destination area candidate No. 3, and the first destination area candidate No. 4 are selected as the destination area candidates C.

  Here, an example of the destination area pattern learning process, the destination address determination process, and the destination area candidate narrowing process using the destination recognition apparatus 100 according to the embodiment of the present invention will be described with reference to FIGS. To do.

  FIG. 8 is a diagram illustrating an example in which the destination area candidate detection processing unit 22 detects the destination area candidate C from the image data P of the delivery object. The example of FIG. 8 is common to all of the destination area pattern learning process, the destination address determination process, and the destination area candidate narrowing process.

  FIG. 8 shows the image data P of the surface of the delivery object including the destination information. Here, a delivery object including, for example, a sender area indicating the address information of the sender and a stamp area on which a stamp is pasted is taken as an example in addition to the destination area indicating the destination information.

  As shown in FIG. 8, first, all the regions including the character string are detected as destination region candidates C. 8 is the destination area candidate C detected in step S101 of the destination area pattern learning process shown in FIG. 2 or step S201 of the destination address determination process shown in FIG. 3, or the destination area candidate narrowing-down process shown in FIG. This is an example of the primary destination area candidate C ′ detected in step S301. In the example of FIG. 8, the area of the destination area candidate C (or the first destination area candidate C ′; hereinafter omitted) is shown surrounded by a broken line.

  Each detected destination area candidate C corresponds to step S102 of the destination area pattern learning process shown in FIG. 2, step S202 of the destination address determination process shown in FIG. 3, or step S302 of the destination area candidate narrowing process shown in FIG. , Serial numbers such as 0, 1, and 2 are assigned to the three destination area candidates C, respectively.

  In this example, the destination area candidate No. 0 is a sender area indicating the address information of the sender, starting with the NG keyword From. The destination area candidate No. 1 is a stamp area in which AIR MAIL is printed on the stamp. The destination area candidate No. 2 is a destination area starting with the specific keyword To.

  FIG. 9 illustrates an example in which the destination region pattern learning processing unit 23 or the destination address determination processing unit 24 generates the positional relationship information R from the destination region candidate C in which the number keyword and the address keyword of the corresponding address are detected. FIG. FIG. 9 is an example common to the destination area pattern learning process and the destination address determination process.

  FIG. 9A shows image data P of the surface of the delivery object including the destination information. Here, an example is a delivery object in which there is only one destination area candidate C and the destination area is written. FIG. 9B is an example of the positional relationship information R generated from the destination area candidate C. This is the same as the destination area pattern A registered in the destination area pattern database 33, which is the type of the positional relationship information R.

  FIG. 9A shows an example of how the number keyword and the address keyword of the corresponding address are detected from the destination area candidate C. That is, FIG. 9A is an example of the number keyword and the address keyword of the corresponding address detected in step S103 to step S105 of the destination area pattern learning process or step S203 of the destination address determination process.

  FIG. 9B is an example of the positional relationship information R generated by the destination area pattern learning processing unit 23 or the destination address determination processing unit 24. FIG. 9B is an example of a destination area pattern A that is a type of the positional relationship information R registered in the destination area pattern database 33.

  In the case of the destination area pattern learning process, the type of the positional relationship information R in FIG. 9B generated from the destination area candidate C having the number keyword in FIG. 9A and the address keyword of the corresponding address is the destination area pattern. A is newly registered (step S110) or updated (step S111).

  For example, when the number keyword input by the operator is “22222”, the ZIP code “22222” is detected from one of the destination area candidates C (the number keyword in FIG. 9A, step S103). Thereafter, the corresponding address “X City, IA USA” of the ZIP code “22222” is acquired from the number keyword / corresponding address dictionary unit 32 (step S104). Subsequently, address keywords “X City”, “IA”, and “USA” are detected (address keyword in FIG. 9A, step S105).

  Thus, the positional relationship information R shown in FIG. 9B can be generated from the number keyword and the three address keywords detected in FIG. 9A (step S107), and the type of the positional relationship information R is If it is not registered in the destination area pattern database 33, a destination area pattern number is allocated and newly registered in the destination area pattern database 33 as the destination area pattern A (step S110). On the other hand, if the type of the positional relationship information R is already registered in the destination area pattern database 33, a predetermined numerical value is added to the score value of the destination area pattern A in the destination area pattern database 33. The contents of the data are updated (step S111).

  Alternatively, in the case of the destination address determination process, the type of the positional relationship information R in FIG. 9B generated from the destination area candidate C having the number keyword in FIG. 9A and the address keyword of the corresponding address is the destination area. This is a target to be checked against the destination area pattern A registered in the pattern database 33 (step S207).

  From one of the destination area candidates C, “22222”, “X City”, “IA”, and “USA” that are the address keywords of the number keyword and the corresponding address are detected (the number keyword and the correspondence in FIG. 9A). Address keyword of address, step S203).

  The positional relationship information R shown in FIG. 9B can be generated from the number keyword and three address keywords detected in FIG. 9A (step S205) and registered in the destination area pattern database 33. The matching destination area pattern A is collated from among the destination area patterns A (step S207).

  According to the embodiment described above, when performing destination recognition of a delivery object, the likelihood of being a destination address based on the destination area pattern database 33 constructed by learning from among a plurality of destination area candidates C. A character string of a destination area candidate C having a high can be selected, and a destination address can be specified with higher accuracy.

  Further, the destination area pattern learning before operation eliminates the need for an operator or greatly reduces the number of operations when determining a destination address during operation, and the address recognition apparatus 100 can be automated.

  Furthermore, the destination recognition apparatus 100 according to the embodiment can be used in the operation target area where the learning is performed in any area if the destination area pattern learning is performed for each destination recognition apparatus 100 in each operation area. .

  An embodiment in which both the destination area pattern learning process and the destination address determination process are executed at the same time is also assumed.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 100 ... Destination recognition apparatus 11 ... Image reading part 12 ... Number keyword input reception part 13 ... Destination address output part 21 ... Image reading control part 22 ... Destination area candidate detection process part 23 ... Destination area pattern learning process part 24 ... Destination address determination Processing unit 31 ... Image storage unit 32 ... Number keyword / corresponding address dictionary unit 33 ... Destination region pattern database 34 ... Matching result storage unit 35 ... Destination region candidate narrowing result storage unit

Claims (16)

An image processing unit that reads an image of the surface of the delivery object and generates image data of the delivery object;
In accordance with the image data generated by the image processing unit, a number keyword input receiving unit that inputs a number keyword that is a number assigned to each address area;
A number keyword / corresponding address dictionary section in which a corresponding address corresponding to the number keyword is stored;
A destination area candidate detection processing unit that detects a destination area candidate that is an area including a character string from the image data;
The number keyword received by the number keyword input receiving unit is detected from the destination region candidates detected by the destination region candidate detection processing unit, and among the number keywords stored in the number keyword / corresponding address dictionary unit, the number The corresponding address corresponding to the number keyword received by the keyword input receiving unit is acquired, and from the acquired corresponding address, an address keyword that is a term constituting the address is detected, and the detected number keyword and the detected address are detected. A destination area pattern learning processing unit that generates first information that is information indicating a positional relationship between keywords;
A destination area pattern database in which the first information generated by the destination area pattern learning processing unit is registered as a destination area pattern;
A destination recognition apparatus comprising: In the destination area pattern database, relative positional relationship information between the number keyword and the address keyword is registered as a destination area pattern.
The destination recognition apparatus according to claim 1. In the destination area pattern database, position information of the number keyword and position information of the address keyword in the destination area candidate are registered as a destination area pattern.
The destination recognition apparatus according to claim 1. An image processing unit that reads an image of the surface of the delivery object and generates image data of the delivery object;
In accordance with the image data generated by the image processing unit, a number keyword input receiving unit that inputs a number keyword that is a number assigned to each address area;
A number keyword / corresponding address dictionary section in which a corresponding address corresponding to the number keyword is stored;
A destination area candidate detection processing unit that detects a destination area candidate that is an area including a character string from the image data;
The number keyword received by the number keyword input receiving unit is detected from the destination region candidates detected by the destination region candidate detection processing unit, and among the number keywords stored in the number keyword / corresponding address dictionary unit, the number The keyword input reception unit acquires a corresponding address corresponding to the number keyword received, detects an address keyword that is a term constituting the address from the acquired corresponding addresses, and suggests that the character string is a destination address A destination area pattern learning processing unit that detects a specific keyword that is a term to be detected and generates second information that is information indicating a positional relationship between the detected specific keyword, the detected number keyword, and the detected address keyword; ,
A destination area pattern database in which the second information generated by the destination area pattern learning processing unit is registered as a destination area pattern;
A destination recognition apparatus comprising: In the destination area pattern database, as a destination area pattern, relative positional relationship information of the specific keyword, the number keyword, and the address keyword is registered.
The destination recognition apparatus according to claim 4. In the destination area pattern database, position information of the specific keyword, position information of the number keyword, and position information of the address keyword in the destination area candidate are registered as a destination area pattern.
The destination recognition apparatus according to claim 4. In the destination area pattern database, the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit is obtained by the destination area pattern learning processing unit. The first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit is compared with the registered destination area pattern registered in the destination area pattern database. If it does not match any of the registered destination area patterns, the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit as a new destination area pattern Information is registered in the destination area pattern database;
The destination recognition apparatus according to any one of claims 1 to 6. In the destination area pattern database, the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit is obtained by the destination area pattern learning processing unit. Matched with the registered destination area pattern registered in the destination area pattern database and matches the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit If there is a registered destination area pattern to be updated, the contents of the matching registered destination area pattern are updated;
The destination recognition apparatus according to any one of claims 1 to 6. From a destination area candidate detected by the destination area candidate detection processing unit, a number keyword and an address keyword are detected, and third information which is information indicating a positional relationship between the detected number keyword and address keyword is generated, The generated third information is collated with a registered destination area pattern registered in the destination area pattern database, and a destination address is selected from all the destination area candidates detected by the destination area candidate detection processing unit. A destination address determination processing unit for selecting a character string of an area candidate;
When there is a registered destination area pattern that matches the third information generated by the destination address determination processing unit as a result of collation by the destination address determination processing unit, the number keyword and address keyword of the matching registered destination area pattern Information indicating the positional relationship is a verification result storage unit that is stored as a verification result;
The destination recognition device according to claim 7 or 8, comprising: The destination address determination processing unit further detects the specific keyword from destination region candidates detected by the destination region candidate detection processing unit, and detects the detected specific keyword, the detected number keyword, and the detected address keyword. 4th information which is the information which shows the positional relationship of this, The said produced | generated 4th information is collated with the registered destination area pattern registered into the said destination area pattern database,
The matching result storage unit, when there is a registered destination area pattern that matches the fourth information generated by the destination address determination processing unit, by matching by the destination address determination processing unit, the matching registered destination area Information indicating the positional relationship between the specific keyword, number keyword, and address keyword of the pattern is stored as a matching result.
The destination recognition apparatus according to claim 9. In the destination area pattern database, the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit is obtained by the destination area pattern learning processing unit. The first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit is compared with the registered destination area pattern registered in the destination area pattern database. If it does not match any of the registered destination area patterns, the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit as a new destination area pattern The information includes the destination area pattern together with a score value corresponding to the number of appearances of the first information or the second information. It is registered in the database down,
The destination recognition apparatus according to any one of claims 1 to 6. In the destination area pattern database, the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit is obtained by the destination area pattern learning processing unit. Matched with the registered destination area pattern registered in the destination area pattern database and matches the first information generated by the destination area pattern learning processing unit or the second information generated by the destination area pattern learning processing unit If there is a registered destination area pattern to be updated, a score value corresponding to the matching registered destination area pattern is updated corresponding to the number of appearances of the destination area pattern;
The destination recognition apparatus according to any one of claims 1 to 6. Number keywords and address keywords are detected from the destination area candidates detected by the destination area candidate detection processing unit, the detected third information is generated, and the generated third information is used as the destination area pattern database. Are compared with the registered destination area pattern registered together with the score value, and out of all the destination area candidates detected by the destination area candidate detection processing unit, based on the score value, A destination address determination processing unit for selecting a character string;
When there is a registered destination area pattern that matches the third information generated by the destination address determination processing unit as a result of collation by the destination address determination processing unit, the number keyword and address keyword of the matching registered destination area pattern Information indicating the positional relationship is stored as a matching result together with the score value of the matching registered destination area pattern;
The destination recognition device according to claim 11, comprising: The destination address determination processing unit further detects the specific keyword from the destination region candidates detected by the destination region candidate detection processing unit, generates the detected fourth information, and generates the generated fourth information. Is compared with a registered destination area pattern registered with the score value in the destination area pattern database,
The matching result storage unit, when there is a registered destination area pattern that matches the fourth information generated by the destination address determination processing unit, by matching by the destination address determination processing unit, the matching registered destination area Information indicating the positional relationship between the specific keyword of the pattern, the number keyword, and the address keyword is stored as a matching result together with the score value of the matching registered destination area pattern.
The destination recognition apparatus according to claim 13. The collation result storage unit
Further, the information indicating the positional relationship between the number keyword and the address keyword of the matching registered destination region pattern, or the information indicating the positional relationship between the specific keyword, the number keyword and the address keyword of the matching registered destination region pattern Information indicating the positional relationship between the number keyword and address keyword of the matching registered destination area pattern based on the score value of the registered destination area pattern to be matched, or the specific keyword, number keyword, and address keyword of the matching registered destination area pattern Stored together with the ranking of information indicating the positional relationship of
The destination recognition apparatus according to claim 13 or 14. The destination address determination processing unit
Further, from the destination area candidate detected by the destination area candidate detection processing unit, an NG keyword that is a term suggesting that the character string is other than the destination address is detected, and the destination area candidate from which the NG keyword is detected is determined. A non-destination area that does not correspond to the destination area
The destination recognition device according to claim 9, claim 10, or claim 13 to claim 15.

Images