JP2001109843A - Method and device for character recognition - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform character recognition corresponding to a cell structure by properly recognizing the cell structure from a document format which has one item represented by more than one cell and one set of data represented by more than one cell. SOLUTION: A character recognition system performs cell extraction for extracting the minimum unit of an area encircled with ruled lines, etc., as a cell from image data obtained by reading an image out of a document, etc., specifies item rows and data rows by arraying respective cells in the row and column directions and also generates table cell structure data showing the structure of the respective cells, the relation between cells, etc., and performs a recognizing process for the cells according to the table cell structure data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition method and apparatus for recognizing a character image and generating character information corresponding to the character image, for example, by reading a form described in a table format. The present invention relates to a character recognition method and apparatus for recognizing a table structure and characters.

[0002]

2. Description of the Related Art In recent years, there has been known a character recognition technique for inputting characters at high speed by reading a character image recorded on paper or the like and outputting a corresponding character code. OCR (Optical Character Rea
der) is constituted by software installed in a general-purpose computer device or a dedicated device, and can recognize various characters. Recent OCRs include handwriting, alphanumeric
In addition to processing Katakana (ANK) characters, kanji, and Western characters as recognition targets, it can also recognize the layout and tabular state of them and generate code data in a format according to the recognition target state. Further, it is desired to accurately recognize recognition targets of various styles.

For example, when performing character recognition processing on a general form in a tabular form surrounded by a ruled line or the like as a general form, the area of the item surrounded by the ruled line is treated as a cell, and Was recognized.
For example, Japanese Patent No. 2740335 discloses a table reader having an automatic cell attribute determination function. According to the invention described in this publication, when reading a character in a cell surrounded by a ruled line, a cell attribute relating to the character content in the cell such as a number, a symbol, and a format is automatically determined, and a character recognition process corresponding to the cell attribute is performed. Is performed to improve the character recognition rate.

[0004]

However, in the table format which can be read accurately by the conventional table reading function, one item is represented by one cell, and data corresponding to the item is directly below the item. Was represented by one cell.

However, there are various document layout structures. For example, when recognizing a character recorded on a form tabulated by a ruled line or the like, an item or data cell is divided into a plurality of cells. It is difficult to accurately perform character recognition in accordance with the table format for a table configured as described above.

For example, in the case of a table format in which an item cell is composed of a plurality of cells, or in the case of a table format in which one data cell is composed of a plurality of cells, it is necessary to accurately determine the table format. Could not. Therefore, for example, each of the closely related areas is processed as an unrelated area. As a result, there is a problem that a line which is originally the same line is recognized as a plurality of lines, or a plurality of cells are recognized despite the same item or one cell content.

The present invention solves the above-mentioned drawbacks of the prior art, and recognizes the cell structure of each tabulated area with high accuracy.
For example, it is possible to appropriately recognize a cell structure from a form in which one item is represented by a plurality of cells or a set of data is represented by a plurality of cells, and perform character recognition processing according to the cell structure. It is an object to provide a character recognition method and apparatus.

[0008]

In order to solve the above-mentioned problems, the present invention recognizes character information corresponding to image information in accordance with image information obtained by reading an image recorded on an information recording medium. In the character recognition method, this method includes inputting image information, extracting a minimum unit cell in a table area surrounded by ruled lines from the image information, and generating coordinate information representing a cell position for each cell. A cell extraction step, a table area alignment step of identifying the structure of the table area by aligning each cell, and generating alignment information representing the structure; and a text from image data corresponding to the cell based on the coordinate information and the alignment information. A table recognizing step including a recognition step of recognizing information and a recognition result outputting step of outputting character information recognized in the recognizing step; The go, characterized in that the alignment information indicative of a relation between the cells each generate for each cell.

In this case, in the table sorting step, when performing row-direction sorting, an item cell representing an item in the table area and an item row by the item cell are specified, and the item row is determined in the column-direction sorting process. The information for specifying the data cell in the data row corresponding to each item cell may be described in the alignment information.

In this case, in the table sorting step, information for specifying a data cell corresponding to each data cell may be described in the sorting information.

The table sorting step includes a step of recognizing the nested state of the item cells in the item row, and a step of creating a composite cell representing the nested item cell according to the recognized nested state. In the step, character information may be recognized for a data cell corresponding to the composite cell.

The table sorting step includes a step of combining data cells in a data row in accordance with the width of the item cell to form a combined cell. The recognizing step includes a step of combining character information for each data cell corresponding to the combined cell. Should be recognized.

The table sorting step includes a step of recognizing the nesting state of the item cells in the item row, and, in accordance with the recognized nesting state, changing a data cell corresponding to the nested item cell in accordance with the width of the item cell. The identifying step may include recognizing character information for the identified data cell.

According to another aspect of the present invention, there is provided a character recognition apparatus for recognizing character information corresponding to image information in accordance with image information obtained by reading an image recorded on an information recording medium. An input unit for inputting image information; and a cell extracting unit for extracting a minimum unit cell in a table area surrounded by a ruled line from the image information and generating coordinate information indicating a cell position for each cell. And a table aligning means for identifying the structure of the table area by aligning the cells and generating alignment information representing the structure, and recognizing character information from image data corresponding to the cells based on the coordinate information and the alignment information. A recognition unit that includes a recognition result output unit that outputs character information recognized in the recognition step, wherein the table alignment unit performs a column direction alignment process and a row direction alignment process. The alignment information indicative of a relation between the cells and generates respectively for each cell.

In this case, the table sorting means specifies the item cells representing the items in the table area and the item rows by the item cells when performing the row direction sorting process, and performs the item row by the column direction sorting process. The information for specifying the data cell in the data row corresponding to each item cell may be described in the alignment information.

In this case, it is preferable that the table sorting means further describes information for specifying the data cell corresponding to each data cell in the sorting information.

The table sorting means includes means for recognizing the nested state of the item cells in the item row, and means for creating a composite cell representing the nested item cells according to the recognized nested state. The means may recognize the character information for the data cell corresponding to the composite cell.

The table sorting means includes means for combining data cells in a data row in accordance with the width of an item cell to create a combined cell, and the recognizing means includes character information for each data cell corresponding to the combined cell. Should be recognized.

The table sorting means includes means for recognizing the nested state of the item cells in the item row, and, in accordance with the recognized nested state, data cells corresponding to the nested item cells in accordance with the width of the item cells. And a recognizing means for recognizing character information for the specified data cell.

[0020]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of a character recognition method and apparatus according to the present invention. Referring to FIG. 3, there is shown a character recognition system 10 according to the present embodiment. The system 10 includes, for example, an image recorded on an information recording medium such as a form 400 shown in a table form as shown in FIG. , And recognizes the table structure and the written characters appropriately, and forms 400 according to the recognition result.
This is a system having a table reading function of generating code data representing the following. In the item row 402 of the form 400 according to the present embodiment, each column of a transfer destination, a deposit type, an account number, a recipient name and an amount is described, and the transfer destination column further includes a bank name and a branch name in two columns. Are divided into Each of these fields is indicated by an area surrounded by a ruled line such as a solid line. In each line below this item line, a data line 404 in which data corresponding to each item is described is arranged surrounded by a ruled line. ing. In this example, the first row of data row 404 contains data ₁₁ to
An area in which data ₁₅ is described is formed, and such data rows are arranged over the last rows of data ₈₁ to data ₈₅ . The system 10 accurately analyzes and recognizes the structure and character information of the table described in the form 400, and generates ruled line codes, character codes, and format information according to the information. In the following description, parts not directly related to the present invention are not shown and described, and reference numerals of signals are represented by reference numerals of connection lines in which the signals appear.

Returning to FIG. 3, the image input device 302 is an image scanner that reads an image printed and printed on a sheet form 400 such as paper and outputs the image data to an output 304. The image input device 302
Optically read characters and figures recorded in 0
B: Generates color or monochrome image information, converts the generated image information into digital image data, and outputs it to an output 304.

An output 304 of the image input device 302 is connected to a computer device 306. The computer device 306 recognizes a character type and a table structure from the input image data and generates code data corresponding to the recognition result. Processing device. Computer device 306 has a keyboard
An input device such as a mouse 308 and a mouse 310, a display device 312 for displaying a processing screen and a processing result, and a printer 314 for printing out desired information are connected to each other.

A facsimile apparatus 316 having a modem for transmitting and receiving image data in a predetermined transmission format is connected to the computer apparatus 306 via a connection line 318, and is connected via a line 320 such as a telephone line and a communication line. The image data transmitted from the destination is received and stored in the computer device 306. The computer device 306
The stored image data 318 is also subjected to recognition processing for tables and characters in the same manner as the output data 304 from the image input device 302.

The computer 306 in this embodiment
Is a general-purpose computer having a function of processing image data input to the input 304 and the input 318 according to a procedure according to an application program, filing the processing result, and displaying and printing out. The computer device 306 is provided with a general-purpose information processing function and a control function for peripheral devices in which a basic program for controlling and transmitting / receiving information to / from each connected device is introduced. Further, a recognition processing program for realizing an image input function, a character recognition function, and a recognition result output function is installed in a storage device (not shown) in the computer device 306, and the computer device 306 stores these programs in a main storage. And performs control and processing according to these programs. This computer device 306
Is recorded on a recording medium such as an optical disk (not shown), a server, or the like, and the program data read therefrom is stored in a storage device in the computer device 306 and is introduced to be usable.

When the facsimile apparatus 316 connected via the connection line 318 is a modem having a facsimile transmission function, the computer apparatus 306 performs facsimile transmission according to the procedure according to the installed communication software. Go and store the received data. Computer equipment
306 decompresses the received data to reproduce the image data, and performs a character and form recognition process on the reproduced image data.

The functional configuration of the computer device 306 will be described with reference to the functional block diagram shown in FIG.
An input unit 200 to which the processing unit 310 is connected and an output unit 202 for outputting a processing result are provided.
Are connected to the control unit 220 via the bus 204.

An image input unit 206 stores image data output from the image input device 302 and the facsimile device 316.
304, 318, and has a function of converting image data into binary data representing a monochrome image in a format compatible with character recognition and a file format. The image input unit 206 outputs the converted binary data to the bus 204 and stores it in the image memory 208. When image data of a binary image is input to the image input unit 206 in advance, the image input unit 20
6 causes the image data to be stored in the image memory 208.

The image memory 208 connected to the bus 204
Multiple forms 400 entered in input 304 or input 318
It has a storage area for storing image data read from the. The image data stored in the image memory 208 is read out to the bus 204 in response to the control of the control unit 220, and the read out image data is used to specify the table cell structure recorded in the form 400. Are input to the cell extracting unit 210, and are input to the recognizing unit 216 in the character recognition stage.

The cell extracting section 210 is a functional section for extracting, from the image data stored in the image memory 208, an area surrounded by ruled lines as cells. The cell extracting unit 210 generates coordinate information indicating the position of the extracted cell, and stores the generated coordinate information in the table cell information memory 214.

More specifically, the cell extracting unit 210 scans the image data in the form 400 in the horizontal (row) direction and the vertical (column) direction to obtain black data continuous in the horizontal direction and black data continuous in the horizontal direction. Is extracted. The cell extraction unit 210 recognizes black data that has continued for a predetermined length in the extracted black data as horizontal and vertical ruled lines, respectively. The cell extraction unit 210 determines each of the minimum unit areas among the areas surrounded by the recognized ruled lines as cells, and calculates coordinate information of each cell. The coordinate information is, for example, position coordinates based on a predetermined position of the image data, and is position information indicating four corners and four points of each cell. The cell extracting unit 210 outputs the coordinate information of each cell extracted in this way to the bus 204 and stores it in the table cell information memory 214.

The table aligning unit 212 is a functional unit that aligns each cell in the row (horizontal) direction and the column (vertical) direction with reference to the position coordinates of each cell, and generates alignment information indicating this alignment form. . The table sorting unit 212 sends the generated sorting information to the bus 204.
And stored in the table cell information memory 214. The detailed functional configuration of the table sorting unit 212 will be described.
Has a line determination function of determining a line in image data.

More specifically, the table arranging unit 212 determines item rows and data rows based on the height of each cell, and creates table cell structure data indicating a table structure, cell relations, and the like. The table cell structure data is created for each cell, and is alignment information in a format in which each information of attribute information, right cell information, lower cell information, and data cell information is described as described later.

The table sorting unit 212 determines the height of the item row based on the maximum value of each cell height in the item row in which the item character is described, and performs a classification process on the item row. For example, the table sorting unit 212 scans the items of each cell in the row direction, and describes the cell on the right side of each target cell in the right cell information of the table cell structure data as the right cell of the target cell. I do.

Specifically, as shown in FIG. 5, in the item line, from the left, “transfer destination”, “bank name”, “deposit item”, “account number”, “recipient name” and “amount” are displayed. In each item order, "item cell" is confirmed and determined as attribute information for each item, and the item cell name existing on the right side of each item is confirmed and determined as right cell information. , Create table cell structure data. Finally, check whether there are any remaining item cells in the item row. In this case, table cell structure data is created for cells of the item “branch name” left after scanning in the horizontal direction. As shown in FIG. 6, the item line is determined so as to include these items as the first line, and the data lines are determined from the second line onward. Note that, in this row-direction cell alignment processing, processing for specifying data cell information indicating the data cell arranged in the next row of each item is not performed, and the data cell information is processed in the column-direction cell alignment processing described below. Specify in. The table sorting unit 212 causes the table cell information memory 214 to temporarily store the table cell structure data created in this way.

The table cell structure data created in this way is configured as shown in FIG. In the present embodiment, as shown in FIG. 11A, the structure of the item cell of “transfer destination” is such that the attribute information is “item cell”, the right cell information is “bank name”, the lower cell information is “none”, Then, the data cell information is created as blank, for example, in a blank space. Also,
The structure of the item cell of "bank name" is as shown in FIG. 3 (b), the attribute information is "item cell", the right cell information is "deposit type", the lower cell information is "branch name", and the data cell Information is not set. Similarly, as for “deposit item”, as shown in FIG. 9C, the right cell information is “account number”, the lower cell information is “none”, and the data cell information is unset.

The table sorting section 212 creates table cell structure data for data cells in each data row in the same manner. In this case, the attribute information of each cell is “data cell”, and the right cell information is, if there is a data cell on the right side,
Describe the cell name. For lower cell information,
In the column direction alignment process described later, if there is another cell below the target data cell in the same data row, the data cell name is set, and if not, "none" is set.

The table sorting section 212 has a function of sorting in the column direction. The table sorting unit 212 reads the table cell structure data created in the row direction from the table cell information memory 214, and performs an operation of determining data cell information in each table cell structure data. Specifically, when setting the unset data cell information for each cell, the table sorting unit 212 sets the data cell information of the item “transfer destination” to “data ₁₁ ” as shown in FIG. 10A, for example. The data cell information of "bank name" is set to "none" as shown in FIG.
As shown in (c), the data cell information of the deposit type is
₁₂ ". For the data cell information in the cells of each data row, if there is a data cell in the next row, describe the cell name. For example, if the cell to be processed is the cell of" data ₁₁ " Describes “data _21. ” The table sorting unit 212 supplies the table cell structure data created in this way to the table cell information memory 214, and updates the table cell structure data in the memory.

Further, the table sorting unit 212 has a function of judging a cell structure in an nested state in which one item is composed of a plurality of cells, and setting a composite cell representing one item cell by the plurality of cells. ing. Specifically, the table aligning unit 212
When one item composed of a plurality of cells corresponds to a data cell of a data row, an area obtained by combining the plurality of item cells is virtually set as a composite cell. For example, the "transfer destination" in the item row does not correspond to the column width of the data row even when it is scanned in the vertical direction, so the column width of "bank name" is added to the column width of "transfer destination". Compare with the column width of the data row. In this case, since the total column width of the item cells matches the column width of the data item therebelow, as shown in FIG. 8, "branch name" is added to "transfer destination" and "bank name". A composite cell representing the added three cells is set. This composite cell is represented in the table cell structure data
These three cells are virtually defined as existing in the upper row. The table sorting unit 212 creates table cell structure data for the composite cell. At that time, the table sorting unit 21
2, as shown in FIG. 9, the lower cell information is virtually arranged as “transfer destination”, and the attribute information is set as “composite cell”, the right cell information is set as “deposit type”, and the data cell information is set as “none”. I do. The table sorting unit 212 supplies the information on the composite cell created in this way to the table cell information memory 214, and additionally describes each content of the composite cell in the table cell structure data.

In this way, the table sorting unit 212 expresses the relevance of each cell by the table cell structure data, and creates table structure data in which the cells are aligned in the row and column directions. In particular, for example, in an item row, a composite cell is created for nested item cells, and a description of lower cell information in the same item row is created in the table cell structure data. Is prevented. Also, since the nested item cell is recognized and the composite cell is set, the nested item is represented by the composite cell, and the relationship between the other nested cell and the next row data cell is clearly defined. Of course, the same processing is performed for a data row, and for example, even when one data row is partially formed of a plurality of cells, by setting lower cell information in the same row,
It is possible to specify the table structure. As described above, in the present embodiment, one item is represented by a plurality of cells, and even when one data cell corresponds to this item, a composite cell that handles a plurality of cells as one cell is set. Then, by creating table cell structure data of a composite cell as shown in FIG. 9, the structure of the table cell can be appropriately defined.

The table cell information memory 214 includes a cell extraction unit 210
And the storage area for storing the table cell structure data of the alignment information created by the table aligning unit 212. When the coordinate information and the alignment information are stored in the table cell information memory 214, the recognizing unit 216 starts a character recognition process for the data line.

The recognition unit 216 recognizes a character string written or printed in each area of the image data stored in the image memory 208 based on the coordinate information and the alignment information stored in the table cell information memory 214. It is a functional unit that performs The recognition unit 216 converts the recognized character string into a corresponding character code, and stores the conversion result in the recognition result storage memory 218.

More specifically, the recognizing unit 216 determines a cell to be processed in accordance with the coordinate information and the alignment information of each cell, and among the image data stored in the image memory 208,
Image data representing a character image arranged at a position corresponding to the target cell is read from the image memory 208, and a character code corresponding to the character image is generated. The recognition unit 216 stores and holds a character pattern serving as a reference for recognizing a character image and information indicating its characteristics as a recognition dictionary.
Image data extracted by extracting a character image corresponding to each cell is compared with character patterns and feature information in a recognition dictionary. As a result of the comparison and collation, the recognition unit 216 determines the priority order of the candidate characters in the order of the character pattern having the highest correlation, and selects an appropriate character using an analysis dictionary storing nouns, contexts, and the like. The recognition unit 216 sequentially stores and holds the selected character string as a candidate character string for each cell,
This is performed for all cells linked by the table structure data relating to the form 400. The recognition unit 216 assigns a character code in a predetermined format to the character thus recognized,
In response to a request from the control unit 220, the data is output to the bus 204, and a ruled line code of a predetermined format representing the table format is created from the table structure data and output to the bus 204. Recognition unit 21
6 outputs to the bus 204 format information indicating the table format and the position and size of characters. The format information based on the character code, the ruled line code, and the control character in the predetermined format is not limited to data in a document format, but may be, for example, binary data in a table format used by spreadsheet software.

When a request for changing candidate characters is received from the control unit 220, the recognition unit 216 corrects the recognition result by replacing the characters corresponding to the request with candidate characters. The recognizing unit 216 has a function of performing a character recognition process on the item name described in the item row in advance when specifying the table structure. However, the present invention is not limited to this, and the item name may be recognized before or after recognizing each character recorded in each cell of the data row. Further, in this embodiment, a description is given of the point that character recognition is performed on data in a cell having a table structure. However, the recognizing unit 216 does not include a ruled line or the like. It also has a function to perform character recognition for characters in the box.

The recognition result storage memory 218 has a storage area for storing a recognition result such as a character code, a ruled line code, and format information of the processing result in the recognition section 216. In FIG. 2, the memory circuits of the image memory 208, the table cell information memory 214, and the recognition result storage memory 218 are shown as memory structures of separate blocks. However, the present invention is not limited to this. Each function may be stored and divided in each storage area of the same type of random access memory constituting the storage.

The control section 220 is a function section having a control function for controlling each function section, a transfer function for transferring information from each section to a desired function section via the bus 204, and various arithmetic processing functions. . For example, when the recognition result for the form 400 and one page is accumulated in the recognition result storage memory 218, the control unit 220
These characters, ruled line codes, and format information are read out to the bus 204 and transferred to the output unit 202. The output unit 202 processes the image data stored in the image memory 208 and supplies the processed image data to the monitor device 312 so that the read image is displayed and the information read from the recognition result storage memory 218 is output to the output destination. The image data for display and the image data for printing are created by converting the data into a data in a format according to. As a result, the table and the characters after the recognition processing are displayed on the display screen of the monitor device 312, and the printer device 314 prints and outputs the information on a predetermined sheet.

The processing operation of the system 10 in the present embodiment with the above configuration will be described.
At 100, when the form 400 is set on the reading unit of the image input device 302, the recorded image on the original is scanned,
An electric signal corresponding to the shading is generated, and the image data is input to the image input unit 206 of the computer device 306. Also, when the image data is faxed,
The facsimile apparatus 316 receives this, and the image data that has undergone processing such as decoding is input to the image input unit 206. The image data input to the image input unit 206 is converted into monochrome binary image data if the image data is gradation data such as a color image, and the converted image data is
It is stored in the image memory 208.

When the image data to be recognized is stored in this way, the process proceeds to step 102, where a cell extraction process is performed. In the cell extracting section 210, continuous black data in the horizontal and vertical directions in the image data stored in the image memory 208 is extracted, and the ruled lines in the form 400 are recognized. Next, the area surrounded by the recognized ruled line is determined as a cell, and coordinate information representing the position of each cell is created. The created coordinate information is stored in the table cell information memory 214 as information indicating the cell position with respect to this image data.

When the cells in the form 400 are extracted in this way, the process proceeds to step 104, where the table arrangement processing is performed. In the row direction alignment processing 106, the height of each cell arranged in the row direction at the upper end of the form 400 is determined, and even when there are a plurality of cells in the row direction,
The maximum height of each cell is recognized as the first item line. When the item line is determined, the recognizing unit 216 performs character recognition on the item character in the item line, and converts the content of each item into code data. Next, the item name is recognized from the left cell in the item row, "item cell" is set as each attribute information from "transfer destination" to "amount", and "branch name" in this item row is further set. , The attribute information is set as “item cell”. When the attribute information in the item row is determined, the table cell structure data in which the attribute information is described for each item cell is stored in the table cell information memory 214. The attribute information is similarly set to “data cell” for each cell in the data row.
Are generated and stored in the table cell information memory 214.

When the alignment processing in the row direction is completed as described above, a column direction alignment processing 108 is performed. This process
At 108, data cell information is recognized for each item cell and data cell. At this time, if the nested cell structure is determined, a composite cell is created, and for example, information on the composite cell as shown in FIG. 9 is created. The created information is supplied to the table cell information memory 214 and is additionally described in the stored table cell structure data.

When the table cell structure data is aligned in relation to each other in this manner, the process proceeds to step 110, where the character recognition processing in the data line is started. In the recognizing unit 216, the cell to be processed is determined based on the coordinate information and the alignment information stored in the table cell information memory 214, and the image data corresponding to the target cell is read from the image memory 208. In the present embodiment, the character image of the image data of the data cell in each data row is collated with the recognition dictionary, and code data corresponding to the character image is generated. Further, ruled lines and format information are created by the recognition unit 216 from the table cell structure data, and these character codes, ruled line codes, and format information are stored in the recognition result storage memory 218 via the bus 204.

Next, proceeding to step 112, the information stored in the recognition result storage memory 218 is transferred to the output unit 202 in response to the control of the control unit 220, converted into an image signal for display, and transmitted to the monitor device 312. Supplied. The operator can confirm the recognition result of the table, the character, and the like displayed on the monitor device 312, and can perform a process of correcting the candidate character by operating the input device if correction or the like is necessary. When the recognition result is determined in this way, an image representing the recognition result is printed as needed, or recorded and stored as a file on an information recording medium.

When continuously processing a plurality of sheets 400 in the same table format, the second sheet 400 is used by using the coordinate information of the analysis result recognized on the first sheet and the table cell structure data of the alignment information. By performing the subsequent recognition processing, the processing can be speeded up. Also, by storing the analysis results, even if a plurality of types of forms are mixed, by selecting and using the table cell structure data of the analysis results according to the types of the forms, the table structure can be improved. The recognition process, that is, the process of creating table cell structure data does not have to be performed every time. This is performed by, for example, associating a form number representing a form type with table cell structure data.

Next, another embodiment of the system 10 will be described with reference to FIG. As shown in FIG.
The first embodiment shown in FIGS. 1 to 3 is different from the first embodiment in that a recognition process is appropriately performed on a form 1100 in which a data cell corresponding to an item “amount” is divided by a plurality of digits. Since the same functional configuration as in the first embodiment may be used, the description is omitted.

As shown in the figure, the form 1100 has items of “bank name” to “amount” arranged in the item line, and in particular, each data line corresponding to the item “amount” includes numerals and symbols. A dotted line and a ruled line, such as a broken line, for dividing each digit are recorded so that characters are entered for each digit.

The cell extracting section 210 in this embodiment extracts each of these divided areas as cells from the image data stored in the image memory 208, and stores the coordinate information of each cell in the table cell information memory 214. Store. In addition, the table sorting unit 212 performs row-wise sorting in the same manner as in the above-described embodiment, and then performs column-wise sorting. In the alignment processing in the column direction according to the present embodiment, data cell information is determined by associating item cells according to the width of data cells.

More specifically, the table sorting unit 212 compares the width of an item cell with the width of a data cell disposed thereunder as shown in FIG. Recognize data cells having a width smaller than the cell width. The table sorting unit 212 sums the widths of the data cells until the recognized data cells have the same width as the item cell width. If the result of adding the plurality of data cells matches the item cell width, the data cell Are combined and recognized as one combined cell 1200, and the combined cell 1200 is made to correspond to the item cell. In this case, the table aligning unit 212
Is created, and stored in the table cell information memory 214.

In the example shown in FIG. 11, the data cell corresponding to "amount" is, for example, each cell of "data (1,6)" to "data (1,11)" in the first row of the data row. , The table sorting unit 212 uses these data cells (1,
6) Create merged cells representing (1,11) and create table cell structure data thereof. In this case, the merged cell 1200 has the attribute information “merged cell”, the right cell information and the lower cell information “none”, and the data cell information is “combined cell (2,6 to 2) , 11) ". In the table cell structure of “amount”, the attribute information is “item cell”, the right cell and the lower cell information are “none”, and the data cell information is “combined cell (1, 6 ~ 1,11) ".

The recognizing unit 216 reads out the image data of the data cell corresponding to the merged cell 1200 from the image memory 208 and performs a character recognition process on the image data corresponding to each cell. In particular, the recognition unit 216 performs a recognition process on the character image written in the merged cell 1200 in the data row corresponding to the item cell of “money”. The recognition result is stored in the recognition result storage memory 218 as a recognition result for the combined cell. In this case, the recognizing unit 216 may perform the character recognition process after performing the vertical ruled line removal process on the image data corresponding to the combined cell.

As described above, in the present embodiment, a plurality of data cells corresponding to the item cell of "money" are divided into a plurality of independent areas and the character recognition processing is not performed. Since the recognition process is performed, a recognition result representing a series of numbers and symbols corresponding to the money amount is output as the recognition result of the data row corresponding to the item cell of the “money” column of a plurality of digits that are closely related in meaning. Can be.

Next, another embodiment of the system 10 will be described with reference to FIG. As shown in FIG.
Regarding the item cell "transfer destination" in the item row, the recognition process is appropriately performed for the form 1300 in which the "bank name" and "branch name" are nested and arranged in the row direction below Unlike the third embodiment, the other components may have the same functional configuration as the above-described embodiment, and thus the description thereof will be omitted.

As shown in the figure, in the form 1300, each item of “transfer destination” to “amount” is arranged in an item row, and further, each item of “bank name” and “branch name” is arranged below the “transfer destination”. Is a form arranged in a nested state.

The table sorting unit 212 in the present embodiment performs row-wise sorting and classifies rows in the same manner as in the above-described embodiment, and then performs column-wise sorting. Specifically, FIG.
As shown in (a) to (c), as an example, the table cell structure data for the item cell of “transfer destination” has attribute information “item cell”, right cell information “deposit type”, and lower cell information "Bank name". In the table cell structure data for the item cell of "bank name", the attribute information is "item cell", the right cell information is "branch name", and the lower cell information is "none". Also,
In the table cell structure data for the item cell of “branch name”, the attribute information is “item cell”, the right cell information is “none”, and the lower cell information is “none”. Note that the contents of the data cell information in the row direction alignment processing are unset as in the above-described embodiment. The alignment process is also performed on each data cell in the same manner as in the above embodiment.

Next, in the alignment process in the column direction, when the process of determining data cell information is performed, a process of recognizing nested item cells and associating nested item cells with data cells is performed. .

More specifically, as shown in FIG. 14, when a plurality of rows of item cells exist in an item row, as shown in FIG.
By comparing the width of the upper item cell with the width of the lower item cell disposed therebelow, an item cell having a width smaller than the item cell width is recognized. The table sorting unit 212 sums the widths of the lower item cells until the recognized item cells have the same width as the upper item cell width, and as a result of summing a plurality of item cells, as a result, has the same width as the upper item cell. If the item cell widths in the lower row match the widths of the data cells, the lower item cell is associated with the data cell.

This associating process is a process for defining data cell information in the table cell structure data.
As shown in (a) to (c), the data cell information in the unset state in the items “transfer destination”, “bank name” and “branch name” are respectively “none”, “data ₁₁ ” and “data”. Set to ₁₂ ". The table arranging unit 212 performs the row and column arranging process in this manner to create table cell structure data as alignment information for specifying the structure of each cell and their relationship, and causes the table cell information memory 214 to store the table cell structure data.

The recognizing unit 216 recognizes characters in the image data corresponding to each item cell and data cell based on the coordinate information and alignment information stored in the table cell information memory 214, and recognizes the recognition result. The result is stored in the result storage memory 218. As described above, in the present embodiment, even when the item cells and the data cells are nested and the nested item cells correspond to the data cells, each of the nested item cells and the data By performing the sorting process that associates the cells, even in the form of an item row such as the form 1300, not only is the "transfer destination" associated with the "bank name" or "branch name", but also the "bank" The data cells of the data rows respectively corresponding to the "name" and "branch name" are specified, and a character recognition process is performed for each corresponding data cell.

As described above, in each of the above embodiments,
A table format with multiple item cells and a single data cell, a table format with one item cell and multiple data cells arranged in the horizontal (row) direction, Is
It is possible to appropriately read a table format in which the cell of the item and the cell of the data are nested, and to accurately generate and output the recognition result according to the table structure.

[0068]

As described above, according to the present invention, table cell structure data for recognizing the cell structure of each region in a table with high accuracy is prepared, and one item can be represented by a plurality of cells. Since a cell structure in a form in which a set of data is represented by a cell can be appropriately recognized, character recognition processing according to the cell structure can be accurately performed.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an operation in a character recognition system according to an embodiment to which the present invention is applied.

FIG. 2 is a block diagram illustrating a functional configuration example of a computer device of the character recognition system.

FIG. 3 is a diagram showing a system configuration example of a character recognition system to which the present invention is applied.

FIG. 4 is a diagram illustrating an example of a form on which character recognition processing is performed in the embodiment.

FIG. 5 is a diagram showing a specific state of an item cell in an item row.

FIG. 6 is a diagram showing a determination state of an item row and a data row.

FIG. 7 is a diagram showing an example of table cell structure data created in a row direction alignment process.

FIG. 8 is a diagram showing a creation state of a composite cell.

FIG. 9 is a diagram showing an example of table cell structure data of a composite cell.

FIG. 10 is a diagram showing an example of table cell structure data created by a column direction alignment process.

FIG. 11 is a diagram illustrating an example of a form on which character recognition processing is performed in the embodiment.

FIG. 12 is a diagram showing an example of creating a merged cell from a data cell corresponding to an item “money”.

FIG. 13 is a diagram illustrating an example of a form on which character recognition processing is performed in the embodiment.

FIG. 14 is a diagram showing the correspondence between nested item cells and data cells.

FIG. 15 is a diagram showing an example of table cell structure data created in the alignment processing.

[Explanation of symbols]

 102 Cell extraction processing 104 Table alignment processing 106 Row alignment processing 108 Column alignment processing 110 Recognition processing 206 Image input unit 208 Image memory 210 Cell extraction unit 212 Table alignment unit 214 Table cell information memory 216 Recognition unit 218 Recognition result storage memory

[Procedure amendment]

[Submission date] March 7, 2000 (200.3.7)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0003

[Correction method] Change

[Correction contents]

For example, when performing character recognition processing on a general form in a tabular form surrounded by a ruled line or the like as a general form, the area of the item surrounded by the ruled line is treated as a cell, and Was recognized.
For example, Japanese Patent Publication No. No. 2740335, table reading equipment is disclosed having an automatic cell attribute determination function. According to the invention described in this publication, when reading a character in a cell surrounded by a ruled line, a cell attribute relating to the character content in the cell such as a number, a symbol, and a format is automatically determined, and a character recognition process corresponding to the cell attribute is performed. Is performed to improve the character recognition rate.

Claims (18)

[Claims] 1. A character recognition method for recognizing character information corresponding to image information according to image information obtained by reading an image recorded on an information recording medium, the method comprising: inputting the image information An input step, a cell extraction step of extracting a minimum unit cell in a table region surrounded by ruled lines from the image information, and generating coordinate information for each cell for each cell, and a structure of the table region. A table alignment step of aligning and specifying each cell and generating alignment information representing the structure; and a recognition step of recognizing the character information from the image data corresponding to the cell based on the coordinate information and the alignment information. And a recognition result output step of outputting character information recognized in the recognition step, wherein the table sorting step performs a row direction sorting process and a column direction sorting process. And generating alignment information indicating a relationship between cells for each cell. 2. The method according to claim 1, wherein in the table sorting step, when performing the row direction sorting process, an item cell representing an item in the table area and an item row by the item cell are specified. A character recognition method, characterized in that, in the alignment processing in the column direction, information for specifying a data cell in a data row corresponding to each item cell of the item row is described in the alignment information. 3. The character recognition method according to claim 2, wherein in the table sorting step, information for specifying a data cell corresponding to each data cell is described in the sorting information. 4. The method according to claim 2, wherein the table sorting step includes: recognizing a nested state of the item cells in the item row; and responsive to the recognized nested state, the nested item cells. And a step of creating a composite cell representing the following. The character recognition method characterized in that the character information is recognized for a data cell corresponding to the composite cell. 5. The method according to claim 2, wherein the table alignment step includes a step of combining data cells in the data row according to a width of the item cell to create a combined cell, and the recognition step. Is a method of recognizing the character information for each data cell corresponding to the merged cell. 6. The method according to claim 2, wherein the table sorting step includes: a step of recognizing a nesting state of an item cell in the item row; and a step of recognizing the nesting state according to the recognized nesting state. Identifying the data cell corresponding to the item cell according to the width of the item cell, wherein the recognizing step recognizes the character information for the identified data cell. 7. A character recognition device for recognizing character information corresponding to image information according to image information obtained by reading an image recorded on an information recording medium, wherein the device inputs the image information. Input means, cell extracting means for extracting a minimum unit cell in a table area surrounded by ruled lines from the image information, and generating coordinate information for each cell for each cell, and a structure of the table area. Table aligning means for aligning and specifying each cell and generating alignment information representing the structure; and recognition means for recognizing the character information from the image data corresponding to the cell based on the coordinate information and the alignment information. And a recognition result output means for outputting character information recognized in the recognition step, wherein the table sorting means performs a column direction sorting process and a row direction sorting process. A character recognition device for generating alignment information representing a relationship between cells for each cell. 8. The apparatus according to claim 7, wherein the table sorting means specifies an item cell representing an item in the table area and an item row by the item cell when performing the row direction alignment processing. The character recognition device is characterized in that, in the alignment processing in the column direction, information for specifying a data cell in a data row corresponding to each item cell of the item row is described in the alignment information. 9. A character recognition apparatus according to claim 8, wherein said table sorting means describes information for specifying a data cell corresponding to each data cell in said sorting information. 10. The apparatus according to claim 8, wherein said table sorting means includes means for recognizing a nested state of an item cell in said item row, and said nested item cell in accordance with said recognized nested state. Means for creating a composite cell representing the character information, wherein the recognizing means recognizes the character information for a data cell corresponding to the composite cell. 11. The apparatus according to claim 8, wherein said table sorting means includes means for combining data cells in said data row according to the width of said item cell to create a combined cell, and said recognizing means. Is a character recognition device for recognizing the character information for each data cell corresponding to the merged cell. 12. The apparatus according to claim 8, wherein said table sorting means includes means for recognizing a nested state of an item cell in said item row, and said nested item cell in accordance with said recognized nested state. Means for specifying a data cell corresponding to the item cell in accordance with the width of the item cell, wherein the recognizing means recognizes the character information for the specified data cell. 13. A computer-readable information recording medium in which a character recognition program for recognizing character information corresponding to image information according to image information obtained by reading an image is recorded, wherein the information storage medium comprises: An input procedure of inputting the image information, a cell extraction procedure of extracting a minimum unit cell in a table area surrounded by ruled lines from the image information, and generating coordinate information representing the position of the cell for each cell, A table alignment procedure for identifying the structure of the table area by aligning the cells and generating alignment information representing the structure; and, based on the coordinate information and the alignment information, the character from the image data corresponding to the cell. A recognition procedure for recognizing information; and a recognition result output procedure for outputting character information recognized in the recognition step. A computer readable information recording medium storing a character recognition program, wherein a character recognition program is generated by performing a direction alignment process and a row direction alignment process to generate alignment information representing a relationship between cells. . 14. The information recording medium according to claim 13, wherein, in the table sorting procedure, when performing the row-wise alignment processing, an item cell representing an item in the table area, and an item row by the item cell. Computer-readable information recording, wherein, in the column direction alignment processing, information specifying a data cell in a data row corresponding to each item cell of the item row is described in the alignment information. Medium. 15. The computer-readable medium according to claim 14, wherein said table alignment procedure describes information specifying a data cell corresponding to each data cell in said alignment information. Information recording medium. 16. The information recording medium according to claim 14, wherein the table sorting step includes a step of recognizing a nested state of an item cell in the item row, and a step of recognizing the nested state according to the recognized nested state. Creating a composite cell representing an item cell, wherein the recognition step recognizes the character information for a data cell corresponding to the composite cell. 17. The information recording medium according to claim 14, wherein the table sorting step includes a step of combining data cells in the data row according to a width of the item cell to create a combined cell. A computer-readable information recording medium, wherein the recognition step recognizes the character information for each data cell corresponding to the merged cell. 18. The information recording medium according to claim 14, wherein the table sorting step includes a step of recognizing a nesting state of an item cell in the item row, and a step of recognizing the nesting state according to the recognized nesting state. Specifying a data cell corresponding to the item cell in accordance with the width of the item cell, wherein the recognizing step recognizes the character information for the specified data cell. Information recording medium.