JP4347677B2 - Form OCR program, method and apparatus - Google Patents

Description

  The present invention relates to a form OCR program, method and apparatus for optically reading a document such as a form as image data and performing character recognition from the read image data.

  An OCR (Optical Characterize Recognition) device that optically reads a form such as a slip or a salary report as image data and recognizes characters written on the form from the read image data (for example, see Patent Document 1 below) is known. It has been. Here, the characters include numbers and symbols. The form has an entry field where individual specific characters for each item are entered for multiple items such as name, beneficiary number, salary income, etc. The name (item name) of each item is printed (preprinted) in advance. The entry fields and item names are divided and arranged by a frame constituted by ruled lines. The information that is finally extracted by the OCR device is the characters entered in the entry column. In the OCR device, it is a major issue to identify this entry column and accurately read the characters in the entry column. Yes.

  The OCR device described in Patent Document 1 prepares logical layout information indicating the relative positional relationship between the item name and the entry field in advance for each form type, and corresponds to a desired form by a form type code. The logical layout information to be selected is selected, and the characters in the entry column are recognized while referring to the selected logical layout information. The logical layout information is the textual representation of the item name and entry field layout on the form. The relative positional relationship between the item name and entry field is the item name frame and entry field surrounding each item name. It is shown by the description order of the text representing

  For example, if there is an entry field to the right of the item name “Name” on the form, the file will have an entry field to the right of the item name box of “Name” “Name” & entry field ”. The description order indicates that there is an entry field to the right of the item name frame, and if there is an entry field below the item name frame, By describing the name “name” and then writing “entry column” under the item name “name” after a line break, it is indicated that there is an entry column below the item name frame. Such information is recorded in the logical layout information for all items of the form.

  The logical layout information does not describe the position of the entry field or the item name frame by the absolute coordinate position, but describes only the relative positional relationship between the two. For this reason, by using the logical layout information for specifying the entry field, it is not necessary to accurately adjust the reading magnification when converting the form into image data, and the reference position of the form image data is deviated from the reference coordinate position. Even in this case, the entry column is not unspecified, so the recognition rate of the entry column is improved, and as a result, the character recognition rate is also improved.

Japanese Patent Laid-Open No. 10-207981

  However, even for a form whose entry items are regulated by a public institution such as a salary payment report, the layout of entry fields and item names differs depending on the manufacturer that creates the form. The OCR device described in Patent Document 1 has a problem that the recognition accuracy is significantly lowered when the layout of items is different. Of course, if logical layout information is created in advance for each form having a different layout and registered, it is possible to prevent the recognition accuracy from being lowered. However, since the number of forms with different layouts is enormous, it is virtually impossible to prepare logical layout information for all forms.

  An object of the present invention is to provide a form OCR program, method, and apparatus that can flexibly cope with forms having different item layouts and prevent a reduction in recognition accuracy.

In the form OCR program of the present invention, for a plurality of items, entry fields and pre-printed item names are arranged, and from the form image obtained by reading the form in which the entry field and the item names are partitioned by ruled lines, In a form OCR program that causes a computer to execute a form OCR process for recognizing characters in the entry field, the OCR process is performed on the entire surface of the form image, and the position of the entry frame that defines the entry field, the item name Recognize the position of the item name frame that surrounds and the character string in these frames, and collect the recognized information as one record for each frame, and whether OCR processing is required again for each item name When the record corresponding to the item name that needs OCR processing again is read with reference to the re-OCR designation information that defines whether or not In addition, the OCR process is performed again from the position of the item name frame included in the read record by referring to the position information of the entry field in which the relative positional relationship with the entry field corresponding to each item name is defined in advance. An entry field specifying process step for specifying an entry field to be used, and using the dictionary data suitable for the attribute of the entry field based on the character attribute information of each item defined in advance, the entry field specifying process step And a partial OCR processing step for partially executing the OCR processing on the entry field specified in (1).

In the form OCR method of the present invention, a form image obtained by reading a form in which entry fields and pre-printed item names are arranged for a plurality of items, and the entry field and the item names are partitioned by ruled lines, respectively. In the form OCR method for recognizing characters in the entry field, OCR processing is performed on the entire surface of the form image, and the position of the entry frame that divides the entry field, the item name frame surrounding the item name, Recognize positions and character strings in these frames, and pre-define whether or not OCR processing is necessary again for each item name, and the entire OCR processing step for collecting the recognized information as one record for each frame With reference to the re-OCR designation information, the record corresponding to the item name that needs the OCR processing again is read out, and relative to the entry field corresponding to each item name While referring to the positional information of the entry fields that predefined location relationship, the entry column specifying process step of identifying the answer column should again OCR processing from the position of the item name frame included in the record read out, predefined Based on the character attribute information of each item, the dictionary data suitable for the attribute of the target entry field is used , and the OCR process is partially executed on the entry field specified in the entry field specifying process step. It consists of a partial OCR processing step.

Further, the form OCR device of the present invention has a form image obtained by reading a form in which entry fields and pre-printed item names are arranged for a plurality of items, and the entry field and the item names are partitioned by ruled lines, respectively. In the form OCR device for recognizing characters in the entry field, the OCR process is executed on the entire surface of the form image, and the position of the entry frame that divides the entry field, the item name frame surrounding the item name, Recognizes the position and character strings in these frames, and pre-defines whether or not the OCR processing is necessary again for each item name, and the entire OCR processing unit that collects the recognized information as one record for each frame With reference to the re-OCR designation information, the record corresponding to the item name that needs OCR processing again is read, and the relative position with the entry column corresponding to each item name is read. With reference to the predefined position information of the entry field and a entry column identifying unit for identifying the answer column to be re-OCR processing from the position of the item name frame included in the record read out, the predefined Partial OCR that partially executes OCR processing on the entry field specified by the entry field specifying processing unit using dictionary data suitable for the attribute of the entry field to be processed based on the character attribute information of the item A form OCR apparatus comprising a processing unit.

In the present invention, for a plurality of items, the entry fields and preprinted item names are arranged, and from the form image obtained by reading the form in which the entry fields and the item names are partitioned by ruled lines, In a form OCR program that causes a computer to execute a form OCR process for recognizing the characters of the form, the OCR process is executed on the entire surface of the form image, and the position of the entry frame that divides the entry field and the item names surrounding the item names Recognize the position of the frames and the character strings in these frames, and collect the recognized information as one record for each frame, and whether or not OCR processing is necessary again for each item name in advance. With reference to the defined re-OCR designation information, the record corresponding to the item name that needs OCR processing again is read out, and for each item name, An entry field that specifies the entry field to be subjected to OCR processing again from the position of the item name frame included in the read record with reference to the position information of the entry field in which the relative positional relationship with the entry field to be defined is defined in advance Using the dictionary data suitable for the attribute of the target entry field based on the character processing information of each item defined in advance, and for the entry field identified in the entry field identification process step Since it consists of partial OCR processing steps that partially execute OCR processing, it is possible to flexibly cope with forms with different item layouts without preparing item layout information for each form in advance. A reduction in accuracy can be prevented.

  A form OCR system 10 shown in FIG. 1 includes a main unit 11, an image scanner 12, and an image data server 13, which are connected by a communication network 14 such as a LAN 14. On the paper feed tray 12a of the image scanner 12, for example, a form 16 such as a salary report is set in units of several hundred sheets. The image scanner 12 scans these forms 16 with a CCD image sensor, and outputs one form image data 35 (see FIG. 2) for one form. The form image data 35 is generated as bitmap data that is a set of pixel data. The image data server 13 includes a data storage device such as an HDD (hard disk drive), and accumulates thousands of form image data output from the image scanner 12.

  The main unit 11 accesses the image data server 13 to read the form image data 35 one by one, and performs OCR processing on the read form image data 35. The character data recognized by the main unit 11 is transferred to, for example, a taxation calculation system and processed.

  The main unit 11 is based on, for example, a personal computer or a workstation and has a form OCR program 28 installed therein, and includes a CPU 21, a RAM 22, an operation unit 24, a display 26, and a hard disk drive (HDD) 27. Each part of the main unit 11 is connected by a data bus 23.

  The CPU 21 executes the operating system to control each part of the main unit 11 and executes the form OCR program 28. The RAM 22 is a working memory used when the CPU 21 executes a program. When the form OCR program 28 is executed, the form OCR program 28 and definition data are loaded into the RAM 22. The operation unit 24 includes an input device such as a keyboard and a mouse, and inputs commands and inputs processing conditions to the CPU 21. On the display 26, an operation screen of the form OCR program 28 is displayed. The HDD 27 is a data storage device and stores an operating system, a form OCR program 28, various definition data 29 described later, dictionary data to be referred to in OCR processing, and the like.

  FIG. 2 is an explanatory diagram of the form 16. In this example, a salary income report will be described as a specific example of the form 16. The form 16 is pre-printed with item names such as “payee”, “address”, “name”, “recipient number”, “reading”, etc. in the outer frame 31, and in the vicinity of each item name. Has an entry field. Each item name and entry field are partitioned by ruled lines. For example, the item names “recipient number” and “payment amount” are divided by frames 32a and 33a, respectively, and a frame 32b adjacent to the right of the item name frame 32a of “recipient number” is a recipient. This is an entry frame that constitutes an entry field in which the number itself is entered. An entry field adjacent to the “payment amount” item name box 33a is also divided by a frame 33b.

  FIG. 3 is a flowchart showing the overall procedure of the form OCR processing of the form OCR program 28. The processing steps of the form OCR program 28 include a form image capturing process, a full OCR process, an entry field specifying process, and a partial OCR process. The entry field specifying process includes an item name frame specifying process and a re-OCR area setting process.

  In the form image capturing process, the form image data 35 is read from the image data server 13 for each form. A full OCR process is performed on the form image data 35.

  In the full OCR process, the OCR process is executed on the entire surface of the form 16, and all the item name frames and entry frames existing in the outer frame 31 are recognized. Recognize character strings. The recognized frame is converted into vector data represented by coordinate information, and the character is converted into text data.

  FIG. 4 is a flowchart showing the procedure of the entire OCR process. The entire OCR process is performed in order from the left end to the right end with the origin O set at the upper left of the outer frame 31 as the starting point, and finally reaches the lower right vertex. In the full screen OCR process, when one frame is detected, the frame information, line information in the frame, and character information in the frame are recognized.

  As shown in FIG. 5A, the recognized information is collected as one recognition information record for each frame, and sequence numbers are assigned to the recognition information records in the order of recognition as frame numbers. The line information includes the number of lines in which characters in the frame are entered and the coordinate information of the lines. The character information in the frame includes the number of characters for each line, the recognized character itself, and the coordinate information of each character. Then, the recognition information records for the entire surface are collected and output as the entire OCR result file 36. The output entire OCR result file 36 is temporarily stored in the work area set in the RAM 22 or the HDD 27.

  FIG. 5B is a more specific explanatory diagram of the contents of the entire OCR result file 36. In the form 16, since the upper left frame is recognized first, “1” is assigned as the frame number. The frame is pre-printed with the characters “Payee”, and if the character information recognizing this preprinted character is correctly recognized, it becomes the recognition character “Payee”. . Since the character recognition rate is not 100%, it may not be recognized correctly. In that case, the misrecognized character information becomes the recognized character as it is. In addition, the number of lines in this frame is 3 lines, such as “Payment” for the first line, “Received” for the second line, and “Ru” for the third line. The number of rows is “3”. As for the coordinates of each character, the XY coordinates of the upper left and lower right are extracted for each character.

  The form OCR program 28 executes OCR processing with reference to the dictionary data 30 (see FIG. 1). The dictionary data 30 includes a system dictionary and a user dictionary. The system dictionary is a dictionary that is used universally regardless of the attributes of multiple characters such as alphanumeric characters, symbols, kana, katakana, and kanji, whereas the user dictionary is dedicated to each character attribute. There are several types for each attribute, such as an alphanumeric user dictionary and a symbol user dictionary. The user dictionary has a character recognition rate of the corresponding attribute much higher than that of the system dictionary, but cannot be used for character recognition of other attributes. The system dictionary and the user dictionary are stored in the HDD 27 and are used as appropriate when the CPU 21 executes the form OCR program 27. In the entire OCR process, since the entire area where a plurality of items having different attributes are mixed is the OCR target area, the system dictionary is selected.

  When the entire OCR process is completed, an item name frame specifying process is executed. In the item name frame specifying process, the recognition information record of the item name frame is specified among all the recognition information records in the entire OCR result file 36 based on the item definition file 37.

  As shown in FIG. 6A, the item definition file 37 includes a plurality of definition records for each item name described in the form 16, and each definition record includes an item name and a re-OCR for each item name. Re-OCR designation information for designating whether or not the item name is necessary, position information in the entry column indicating the relative positional relationship between each item name and the entry column corresponding to each item name, and item names in the vicinity of each item name Relative position information. Although not shown, the item definition file 37 includes a re-OCR item number that is an item number of an area in which OCR processing is executed again, as will be described later.

  The position information of the entry column is information indicating in which direction of each item name the entry column is adjacent. The position information in this entry field is defined by numbers. As shown in FIG. 6 (B), “1” is “the box located to the right of the item name is the entry field”, The meaning of “2” is defined as “the frame positioned under the item name is an entry field”. For example, since an entry column is located under the item name for the item of payment amount, the position information of the entry column is designated as “2”.

  Specifically, the relative position information with the nearby item name is such that the item name “name” is on the left of the item name “recipient number”, and the item name “phonetic” is below. It is described by. The relative position information with the neighboring item names is used when an item name frame corresponding to a certain item name cannot be specified in the item name frame specifying process as described later.

  The flowchart shown in FIG. 7 shows a specific procedure of the item name frame specifying process. First, one item name for which the item name frame is to be specified is read from the item definition file 37, and the item name and the recognition character included in the full-scale OCR result file 36 are collated, thereby the item corresponding to the item name. Search for the frame number of the name frame. If the recognized character matches the item name, the frame number is read from the full OCR result file 36 and added to the corresponding item name record in the item definition file 37 as shown in FIG. Thus, the item name frame specifying data file 38 is generated.

  For example, when the item name frame of “recipient number” is specified, “recipient number” is read from the item definition file 37, and this “recipient number” is used as a key in the entire OCR result data file 36. By matching with the recognized character, the frame number of the item name frame to be specified is searched. If the search can be performed, that is, if the character string “recipient number” exists in the full OCR result data file 36, the frame number corresponding to the recognized character is stored in the record of the recipient number in the item definition file 36. In addition, the item name frame specifying data file 38 is created.

  However, if the character string is misrecognized or cannot be recognized in the full OCR process, the character string “recipient number” does not exist in the full OCR result file 36 as a matter of course. When the frame number of the item name frame to be specified cannot be searched in this way, the search process is executed using the item name located in the vicinity of the item name to be specified as a key. For example, when the character string “recipient number” does not exist, the relative position information in the item definition data file 37 is referred to, and a position in the vicinity of “recipient number” such as “name” or “phonetic” is displayed. The item name to be checked is checked, and search processing is executed using the item name as a key. When the character string “name” is found in the entire OCR result data file 36, based on the relative position information (“receiver number” exists on the right side of “name”), The frame number of the item name frame of “number” is estimated. The estimated frame number is added to the item definition data as the frame number of the item name frame to be searched.

  As described above, when the item name frame corresponding to the desired item name cannot be specified, the item name frame is estimated using the item name located in the vicinity of the desired item name as a key, Even when the character string is misrecognized or cannot be recognized in the entire OCR process, a desired item name frame can be specified.

  Such a search based on item names in the vicinity may be executed over all data in the full OCR result file 36. For example, when specifying the item name frame of “recipient number”, first, a search is performed using “recipient number” as a key, and if the search is impossible, the “name” and “ If an item name such as “Reading” is used as a search key and still cannot be found, a search is performed across all data, such as searching using the item name in the vicinity of “Name” or “Reading”. It is also possible. However, this increases the load of the search process, and the search time becomes very long.

  Therefore, the form OCR program 28 limits the search range by defining the search range in the estimation process. The bands (bands 1 to 5) shown in FIG. 2 indicate search ranges, respectively, and the item definition file 37 (see FIG. 6) includes a band number indicating which band each item belongs to. The band is defined as a band-shaped area extending in the horizontal direction since the writing direction of the form 16, that is, in this example, the form 16 is written horizontally. In this example, the width of the “payee” item at the top left is band 1 and “type”, “payment amount”, “amount after deduction of salary income”, “total amount of income deduction” , “Band 2” is defined as the width in which each item of “withholding amount” is arranged. This band is not defined by physical position information such as coordinate information, but is logically defined by item names. That is, the definition of band 1 is not performed by defining the range of band 1 by coordinate information, but by describing a plurality of item names (name, reading, etc.) included in band 1.

  In this way, by limiting the search range, for example, when specifying the item name frame “type”, the search key used for the estimation process is “payment amount”, “amount after deduction of salary income” ”,“ Total amount of deduction for income ”, and“ Withholding amount ”. This reduces the load of search processing and shortens the search time.

  Further, if a desired item name frame cannot be specified even if the search process is executed using a nearby item name as a key, an error occurs. In the case of an error, the frame number field of the item name frame specifying data file 38 is blank. By such an item name frame specifying process, the record of each item name in the item definition file 37 is associated with each record in the entire OCR result file 36.

  After the item name frame specifying process, a re-OCR area setting process is executed. The re-OCR area setting process outputs a re-OCR area data file 41 for designating an area to be re-OCR based on the item name frame specifying data file 38 and the re-OCR item definition file 39.

  As shown in FIG. 9, the re-OCR item definition file 39 is a file in which information related to the entry column of items to be OCR again is defined for each item. The re-OCR item includes, for example, “recipient number”, “phonetic”, “type”, “payment amount”, “amount after deduction of salary income”, “total amount of income deduction”, “withholding amount” Etc. These items are designated for re-OCR in the item definition file 37. In the re-OCR item definition file 39, information including a re-OCR item number, an item name, and an attribute is one record for each item. The attribute information is information on character attributes of items to be entered such as English letters, numbers, kana, kana and kanji. When the re-OCR process is performed, a user dictionary that matches the attribute is selected based on the attribute information.

  FIG. 10 is a flowchart showing the procedure of the re-OCR area setting process. In the re-OCR area setting process, first, one item name frame record for which re-OCR is specified is read from the item name frame specifying data file 38. Then, based on the position information of the entry field corresponding to the item name frame, the recognition information record of the entry field is identified and extracted from among the recognition information records in the entire OCR result file 36.

  Then, the attribute information of the corresponding item is read from the re-OCR item definition file 39, and unnecessary recognition character information in the entry column is removed based on the attribute information. For example, as shown in FIG. 11, the payment amount entry field 46 includes a character string “5,600,000” indicating the amount in the field, and “inside” and “yen” in the line above it. "Is preprinted. The recognition information record of the entire OCR result file 36 stores all the frame information in the entry field 46 and the line information and character information in the frame.

  The area where the re-OCR processing is to be performed is only the area in which the amount of money (“5,600,000”) is described in the entry field 46. Other character strings need not be recognized, but may cause noise when recognizing necessary character strings. For this reason, the form OCR program 28 removes information on such unnecessary characters from the recognition information record in the re-OCR area setting process. The determination of whether or not the character is an unnecessary character is made based on the attribute information. For example, since the attribute of the payment amount is a number, a character other than a number is determined as an unnecessary character from the character string. When the unnecessary character is removed, the frame information of the entry field 46 and the character coordinates 46a indicating the position of the amount of money in the field remain in the recognition information record.

  Data obtained by combining the recognition information record from which unnecessary recognition character information is removed in this way and the attribute information in the entry column is output as re-OCR area data. Such processing is performed for all items for which re-OCR processing is executed, and a re-OCR area data file 41 in which data for all items are collected is generated.

  In this way, by performing the item frame specifying process and the re-OCR area setting process, the entry field to be OCR again is specified.

  The partial OCR process refers to the re-OCR area data file 41 and partially executes the OCR process for the area of the form image designated as re-OCR. First, one item of area information in the entry column is read from the re-OCR area data file 41. Next, a user dictionary corresponding to the attribute of the read item is set. For example, in the payment amount entry field, the attribute is a number, so a user dictionary for numbers is set. Then, the re-OCR area is identified from the coordinate information of the area information, and the OCR process for that area is executed. This process is repeated for all items. The recognized character string is output to the partial OCR result file 42 together with the re-OCR item number and the item name. In this way, partial OCR processing is performed using a user dictionary corresponding to the character attribute, so that character recognition with high accuracy can be performed.

  Hereinafter, the operation of the above configuration will be described. When the operator sets the form 16 in the image scanner 12 and gives a reading instruction, the image scanner 12 converts the form 16 into image data, and the image data is stored in the image data server 13. Next, when the operator gives a form OCR process execution instruction from the main unit 11, the form OCR program 28 is activated. The form OCR program takes the form image data from the image data server 13 one file at a time and executes the form OCR process.

  First, the entire OCR process is executed to recognize item name frames, entry frames, and character strings in the frames for all items in the outer frame 31, and these recognition information are grouped into recognition information records for each frame. The entire OCR result file 36 is output. By this entire OCR process, the frame layout of each item of the form 16 is recognized.

  Next, an item name frame specifying process is executed based on the entire OCR result file 36 and the item definition file 37 to specify an item name frame. In this item name frame identification process, the corresponding item name frame is identified using the desired item name as a key. If the item name frame cannot be identified, the desired item name is identified using the neighboring item name as a key. Name slots are estimated. For this reason, in the full-screen OCR process, even when a desired item name is erroneously recognized or cannot be recognized, the item name frame can be specified, so that the specification rate of the entry column is improved. The result of the item name frame specifying process is output as an item name frame specifying data file 38.

  Based on the item name frame specifying data file 38 and the re-OCR item definition file 39, a re-OCR area setting process is executed to specify a re-OCR area (an entry field to be OCR again). By this re-OCR area setting process, the position of the entry field and its attribute are specified, and unnecessary character information is removed, and as a result, the re-OCR area data file 41 is output.

  In the partial OCR process, based on the re-OCR area data file 41, the OCR process is executed again for the designated entry field. In this partial OCR process, a user dictionary suitable for the target entry field is selected based on the attribute information, so that highly accurate character recognition is possible.

  As described above, the form OCR program 28 first recognizes the layout of each item of the form 16 by the full OCR process, and then specifies the entry field based on the result information and the item definition data. For this reason, even if the layout of the items is different, it is only necessary to prepare item definition data including necessary item names without preparing layout information for each form in advance. No decrease.

  In the above-described embodiment, a form OCR apparatus in which a form OCR program is installed on the basis of a general-purpose personal computer or workstation is used as the main unit of the form OCR system. Although explained in the example realized by executing, of course, as the main unit, a dedicated form OCR device having a processing unit for executing at least a part of each processing step by dedicated hardware is used. May be.

It is a whole block diagram of a form OCR system. It is explanatory drawing of a form. It is a flowchart which shows the procedure of the whole form OCR process. It is a flowchart which shows the procedure of a whole surface OCR process. It is explanatory drawing of a whole surface OCR result file. It is explanatory drawing of an item definition file. It is a flowchart which shows the procedure of an item name frame specific process. It is explanatory drawing of an item name frame specific process. It is explanatory drawing of a re-OCR area setting process. It is a flowchart which shows the procedure of a re-OCR area setting process. It is explanatory drawing of an unnecessary character removal process. It is a flowchart which shows the procedure of a partial OCR process.

Explanation of symbols

10 Form OCR system 11 Main unit 16 Form 21 CPU
22 RAM
27 HDD
35 Form image data 36 Full OCR result file 37 Item definition file 38 Item name frame specific data file 39 Re-OCR item definition file 41 Re-OCR area data file 42 Partial OCR result file

Claims (3)

For a plurality of items, the entry fields and preprinted item names are arranged, and characters in the entry fields are recognized from a form image obtained by reading a form in which the entry fields and the item names are partitioned by ruled lines. In a form OCR program that causes a computer to execute form OCR processing
OCR processing is performed on the entire surface of the form image to recognize the position of the entry frame that divides the entry field, the position of the item name frame surrounding the item name, and the character string in these frames. Complete OCR processing step to collect the information as one record per frame,
With reference to re-OCR designation information that defines in advance whether or not OCR processing is necessary again for each item name, the record corresponding to the item name that requires OCR processing is read out again, and corresponding to each item name Specify the entry field to identify the entry field to be subjected to the OCR process again from the position of the item name frame included in the read record with reference to the position information of the entry field in which the relative positional relationship with the entry field is defined in advance. Processing steps;
Based on the character attribute information of each item defined in advance, the dictionary data suitable for the attribute of the target entry field is used , and the OCR process is partially performed on the entry field specified in the entry field specifying process step. A form OCR program characterized by comprising a partial OCR processing step for executing. For a plurality of items, the entry fields and preprinted item names are arranged, and characters in the entry fields are recognized from a form image obtained by reading a form in which the entry fields and the item names are partitioned by ruled lines. In the form OCR method
OCR processing is performed on the entire surface of the form image to recognize the position of the entry frame that divides the entry field, the position of the item name frame surrounding the item name, and the character string in these frames. Complete OCR processing step to collect the information as one record per frame,
With reference to re-OCR designation information that defines in advance whether or not OCR processing is necessary again for each item name, the record corresponding to the item name that requires OCR processing is read out again, and corresponding to each item name Specify the entry field to identify the entry field to be subjected to the OCR process again from the position of the item name frame included in the read record with reference to the position information of the entry field in which the relative positional relationship with the entry field is defined in advance. Processing steps;
Based on the character attribute information of each item defined in advance, the dictionary data suitable for the attribute of the target entry field is used , and the OCR process is partially performed on the entry field specified in the entry field specifying process step. A form OCR method characterized by comprising: a partial OCR processing step for executing For a plurality of items, the entry fields and preprinted item names are arranged, and characters in the entry fields are recognized from a form image obtained by reading a form in which the entry fields and the item names are partitioned by ruled lines. In the form OCR device
OCR processing is performed on the entire surface of the form image to recognize the position of the entry frame that divides the entry field, the position of the item name frame surrounding the item name, and the character string in these frames. A full-screen OCR processing unit that collects the information as one record for each frame,
With reference to re-OCR designation information that defines in advance whether or not OCR processing is necessary again for each item name, the record corresponding to the item name that requires OCR processing is read again, and corresponding to each item name Specify the entry field to identify the entry field to be subjected to the OCR process again from the position of the item name frame included in the read record with reference to the position information of the entry field in which the relative positional relationship with the entry field is defined in advance. A processing unit;
Based on the character attribute information of each item defined in advance, the dictionary data suitable for the attribute of the target entry field is used , and the OCR process is partially performed on the entry field specified by the entry field specifying processing unit. A form OCR apparatus comprising a partial OCR processing unit for executing

Images