JPH0757040A - Filing device provided with ocr - Google Patents

Abstract

PURPOSE:To improve the recognition performance of a recognition dictionary for character recognition by performing a learning process at all times on the basis of the recognition results of characters stored in a file device. CONSTITUTION:This device provided with an image input part 2A, a document structure analytic part 4 which analyzes an input image and separates it into areas differing in attribute, a character recognition part 5 which performs character recognition by using the recognition dictionary 9 and the file device 7 is provided with a document analytic process part 24 which performs a document analytic process for character data in a character area stored in the file device by using a word dictionary and grammar; and the character recognition part 5 updates the recognition dictionary 9 on the basis of information regarding misrecognized characters to optimize the recognition dictionary. Further, the document analytic process and recognition dictionary updating process are executed as background processes at all times.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a filing device with OCR. This device, for example, reads a document such as a newspaper clipping with a scanner, the input image is a character area,
This is a device having a function of separately storing in a file device into regions having different attributes such as a graphic region and a photograph region, performing character recognition for a character region, and also storing recognition result data in the file device.

Further, in this apparatus, the data accumulated in the file apparatus can be re-edited by performing a keyword search or the like.

[0003]

2. Description of the Related Art FIG. 9 is an explanatory view of the prior art.
Medium, 1 is a filing device with OCR, 2 is a scanner,
3 is an image memory, 4 is a document structure analysis unit, 5 is a character recognition unit, 7 is a file device, 8 is a keyword search unit, 9 is a recognition dictionary, and 10 is a display.

Conventionally, for example, a device shown in FIG. 9 has been known as a filing device with OCR.
As shown in the figure, the filing device 1 with OCR is provided with a scanner 2, an image memory 3, a document structure analysis part 4, a character recognition part 5, a file device 7, a keyword search part 8, a display 10 and the like. Further, the character recognition unit 5 is provided with a recognition dictionary 9 used in the character recognition process.

In this apparatus, for example, a document such as a newspaper clipping is input, and from the input image (image data), a "text area" such as a text or a headline, a "chart area",
"Photo area" is extracted, automatically separated, and stored in the file device. Further, character recognition processing is performed on the “character area”, and the character data is also stored in the file device. Specifically, it is as follows.

First, the document is read by the scanner 2, the input image (image data) is binarized, and stored in the image memory 3. Next, the document structure analysis unit 4 automatically analyzes the input image stored in the image memory 3 to identify and automatically separate the “character area”, the “graph area”, and the “photo area”. The file is stored (filed) in the file device 7.

The separated "character area" is sent to the character recognition section 5 for character recognition. In this case, the character recognition unit 5 uses the recognition dictionary 9 to perform character recognition processing on the “character area”, and also stores (filing) data (character code) of the recognition result in the file device 7.

When storing the data in the file device 7, the attribute data and the like (text, heading, chart, photograph, etc.) for each area obtained when the document structure analyzing section 4 performs the document analyzing process. ) Is also stored together.

By inputting a keyword from the keyword search section 8 to the data stored in the file device 7 in this way, it is possible to search for character data by the keyword. If there is matching data in this keyword search, the image information, attribute data, etc. are displayed on the display 10.

[0010]

SUMMARY OF THE INVENTION The above-mentioned conventional devices have the following problems. The character recognition function for generating character data to be searched does not perform perfect recognition, and usually causes an error of about several percent. Therefore, although the original information includes the keyword, the search may fail and the search may not be performed.

The present invention solves such a conventional problem by constantly learning the recognition dictionary used for the character recognition processing based on the recognition result of the characters stored in the file device. , It aims at improving the recognition performance of a recognition dictionary.

[0012]

FIG. 1 is a diagram for explaining the principle of the present invention. In FIG. 1, the same parts as those in FIG. 9 are designated by the same reference numerals. 2A is an image input unit, 24 is a sentence analysis processing unit, and 25 is a word dictionary / grammar unit.

The present invention has the following structure to solve the above problems. : An image input unit 2A for inputting a document image, and a document structure analysis unit 4 for analyzing the input document image and separating it into regions having different attributes (character region, chart region, photo region) and extracting various information. A character recognition unit 5 that performs character recognition using a recognition dictionary 9 on a character region among the separated regions, and document data (image data, character data, etc.) obtained by each unit are stored. OCR with file device 7
In the attached filing device, a sentence analysis process is performed for the character data in the character area of the document data stored in the file device 7, the misrecognized character is extracted, and the correct analysis character information is obtained. The unit 24 is provided, and the character recognition unit 5 is based on the information on the misrecognized characters.
A filing apparatus with OCR that enables optimization of the recognition dictionary by updating the recognition dictionary 9.

In the configuration, a word dictionary and a word dictionary / grammar section 25 storing grammatical information are provided, and the sentence by the sentence analysis processing section 24 using the word dictionary / grammar of the word dictionary / grammar section 25. A filing device with OCR for analysis processing.

In the configuration, when the recognition dictionary updating process is performed by the character recognition unit 5, the image data corresponding to the erroneously recognized character is cut out based on the information about the erroneously recognized character to obtain the feature amount, and the feature amount is calculated. A filing device with OCR for updating the recognition dictionary 9 based on the above.

In the configuration, by making the internal processing multitasking, the text analysis processing by the text analysis processing section 24 and the recognition dictionary update processing by the character recognition section 5 are always performed regardless of other processing. A filing device with OCR that can be executed as a background job.

[0017]

The operation of the present invention based on the above configuration will be described with reference to FIG. : When document data is input and stored in the file device, the following processing is performed.

First, a document is read by the image input section 2A,
Enter the image. The input image is temporarily stored in the image memory 3, and then the document structure analysis unit 4 performs a document analysis process.
At this time, the document structure analysis unit 4 scans the data on the image memory 3 to extract and separate the “character area”, the “graph area” and the “photo area”. Further, for each area, information such as the number of areas, the position, the width, and the height is extracted. Then, these data are transferred to and stored in the file device 7.

When it is determined that the character area is a character area in the above processing, the character recognition unit 5 uses the recognition dictionary 9 to perform character recognition processing, and the recognition result data is also transferred to the file device 7. Store. In this way, the image data and the character data are stored in the file device 7.

Description of Optimization Processing of Recognition Dictionary by Background Job The optimization processing of the recognition dictionary by the background job is performed as follows.

First, the text analysis processing section 24 performs text analysis processing on the character data in the character area stored in the file device 7. In this case, the sentence analysis processing unit 24 uses the word dictionary and the grammar of the word dictionary / grammar unit 25 to perform a sentence analysis process to extract misrecognized characters and obtain correct answer information of the characters. , To the character recognition unit 5.

Thereafter, the character recognition section 5 cuts out the image data corresponding to the erroneously recognized character from the file device 7 and converts it into a feature amount. The character recognition unit 5 updates the recognition dictionary 9 based on the converted feature quantity and the category feature quantity stored in the recognition dictionary 9 so that the dictionary is optimal for recognition.

By constantly performing the above processing as a background job, the recognition dictionary can be learned and the recognition dictionary can be optimized. as a result,
The recognition performance of the recognition dictionary can be improved.

[0024]

Embodiments of the present invention will be described below with reference to the drawings. 2 to 8 are views showing an embodiment of the present invention. In FIGS. 2 to 8, the same parts as those in FIGS. 1 and 9 are designated by the same reference numerals.

Further, 11 is a CPU (central processing unit), 1
Reference numerals 2 and 13 denote input / output control units (hereinafter simply referred to as “I / O”), 14 denotes a memory, 17 denotes an optical disk device input / output control unit (hereinafter simply referred to as “optical disk device I / O”),
Reference numeral 18 is an optical disk device, 21 is an image processing unit, 22 is a matching unit (or collating unit), and 23 is a keyword recording / searching unit.

§1: Description of the configuration of the filing device with OCR ... See FIG. 2. FIG. 2 is a device configuration diagram of the embodiment. As shown, OC
The filing device 1 with R includes a CPU 11 and an I / O 1
2, 13, scanner 2, image memory 3, display 10, memory 14, optical disk device I / O 17, optical disk device 18, document structure analysis unit 4, character recognition unit 5, keyword recording / search unit 23, sentence analysis processing unit 24. A word dictionary / grammar unit 25 is provided.

The character recognition section 5 is provided with an image processing section 21, a matching section 22 and a recognition dictionary 9. Functions and the like of the respective units are as follows. (1): The CPU 11 is a processor that performs various controls in the filing apparatus with OCR.

(2): The I / Os 12, 13 are the scanner 2,
And each input / output control of the display. (3): The scanner 2 optically reads a document and inputs a document image.

(4): The display 10 displays various information. (5): The optical disc device I / O 17 is the optical disc device 1
8 for input / output control.

(6): The optical disk device 18 is a file device (for example, a magneto-optical disk device) capable of recording and reproducing data. (7): The document structure analysis unit 4 analyzes the input image and divides it into each area such as "character area", "chart area", "photo area", or detects the coordinates of each area. (Details will be described later).

(8): The image processing section 21 performs processing of image data (image data) when performing character recognition processing in the character recognition section 5, for example, feature extraction processing.

(9): The matching unit 22 is the image processing unit 21.
The character recognition processing is performed by comparing the information (feature amount, etc.) processed in (1) with the recognition dictionary 9. (10): The keyword recording / retrieval unit 23 records a keyword of character data in the optical disc device 18 or targets character data stored in the optical disc device 18,
It is something that searches for keywords.

(11): The sentence analysis processing section 24 uses the word dictionary / grammar section 2 for the character data in the optical disk device 18.
The sentence analysis processing is performed using the information of 5 (details will be described later).

(12): The word dictionary / grammar section 25 stores a word dictionary and grammatical information necessary for sentence analysis. (13): The memory 14 is a work memory.

§2: Description of data in the optical disk device
.. Refer to FIG. 2. As the data to be stored in the optical disc device 18, for example ,: file number ,: image information ,: total area number,
: Number of image areas ,: number of character areas, etc.

The above: The image information is information such as "width", "height", etc., which is composed of image data, binary code string, etc .: The number of image areas is "image area number",
It consists of "image area start position", "width", "height", and the like.

The above: The number of character areas is composed of "character area contents" and "attributes", and "attributes" are "vertical writing / horizontal writing", "area start position", "width", and "height". "
It includes information such as "relationship with image area".

§3: Description of Document Data Saving Procedure--See FIG. 2 Hereinafter, the document data saving procedure will be described with reference to FIG. (1): First, when the apparatus is activated, the CPU 11 activates the scanner 2 to read a document and start inputting an image. The input image (image data) is binarized and then temporarily stored in the image memory 3.

(2): Next, the CPU 11 activates the document structure analysis unit 4 and analyzes the input image (image data) in the image memory 3. At this time, the document structure analysis unit 4
Scan the data in the image memory 3 to display "character area"
And "Chart area" and "Photo area" are extracted and separated. Further, for each area, information on the number of areas, the position, the width, and the height is extracted.

Under the control of the CPU 11, these data are transferred to the optical disk device 18 and stored in the medium (optical disk). In this case, for the “character area”, it is automatically determined whether the content is “vertical writing” or “horizontal writing”, and the information is also stored in the optical disc device 18.

Further, with respect to the character area on the side of a photograph or a chart, the attribute of the explanatory note of the photograph or the chart is determined by, for example, histogram calculation and determination by a threshold value, and transferred to the optical disc device 18 to be a medium. (Optical disc).

That is, the "graph / table area" and the "photograph area"
The relationship between the "image area" and the "character area" is automatically determined (the character area is sufficiently smaller than the size of the image area when the distance between the image area and the character area is within a certain threshold). For the “character area” that is determined to be the description of the “image area”, the information is also stored in the optical disc device 18.

When it is determined in the above process that the character region is a character region, the character recognition unit 5 performs a character recognition process on the corresponding region and stores it in the optical disk device 18. (3): In the analysis processing by the document structure analysis unit 4,
For the area determined to be the “character area”, the CPU 11
Activates the character recognition unit 5 to perform character recognition processing.

In this case, in the character recognition section 5, the image processing section 21 performs image processing (feature extraction, etc.) only on the range of the area determined to be the "character area" on the image memory 3, and the matching section 22 uses the recognition dictionary 9 to perform character recognition processing.

(4): When the character recognition processing by the character recognition unit 5 is completed, the character data (character code, etc.) of the recognition result is stored in the optical disk device 18 together with the attributes obtained by the document structure analysis unit 4. To do. In this case, in the optical disc device 18, the “image data” and the “character data” are 2
It is managed as one pair.

(5): For example, the above-mentioned processes are processed for each page of the document to be read and repeated until the unit of the document designated by the operator is completed. §4: Description of keyword search procedure--see FIG. 2 When performing a keyword search on the data stored in the optical disk device, the following steps are performed.

(1): First, an input means such as a keyboard is used to input a keyword that is considered to be contained in the document data to be searched. (2): The keyword recording / search unit 23 performs a full-text search on all the character data portions stored in the optical disk device 18 based on the input keyword. At this time, if there is a keyword in the character data portion, only that file is targeted for the subsequent search.

(3): The above-mentioned processing is repeated to narrow down the candidates of the file to be searched from a large number of files, and when the number of the candidates is reduced, the image data corresponding to the character data is extracted by the operator's instruction. To do.

Then, under the control of the CPU 11, the search result data is sent to the display 10 and displayed. The operator looks at this display screen and confirms whether or not the file is the target file.

(4): If the file is the target file,
Print out, transfer data to another disk device, FA
The data is output by means such as X transmission. §5: Description of document structure search procedure-See Fig. 2 (1): First, the operator selects one of the displayed document structure keywords or inputs a key, and starts the search using it. To do. At this time, the CPU 11 activates the keyword recording / search unit 23.

(2): The keyword recording / retrieving unit 23 retrieves the character data of all the documents stored in the optical disk device 18, and only the documents including the document structure keyword are targeted for the following.

(3): The above process is repeated to narrow down the candidates of the file to be searched from a large number of files. When the number of candidates is reduced, the image data corresponding to the character data is extracted according to the operator's instruction.

Then, under the control of the CPU 11, the search result data is sent to the display 10 and displayed. The operator looks at this display screen and confirms whether or not the file is the target file.

(4): If the file is the target file,
Print out, transfer data to another disk device, FA
The data is output by means such as X transmission. §6: Description of dictionary optimization process (automatic dictionary update process) by background job ... See FIG. 2 The recognition dictionary optimization process (automatic dictionary update process) by the background job is performed as follows. .

(1): The sentence analysis processing unit 2 by the CPU 11
4 is activated, the sentence analysis processing section 24 reads out the character data in the character data area of the file stored in the optical disk device 18 and stores it in the memory 14.

(2): At this time, the image data representing the character area of the image information of the file is stored in the memory 14. (3): The sentence analysis processing unit 24 performs sentence analysis processing on the character data in the memory 14 using the word dictionary of the word dictionary / grammar unit 25 and the grammar, and extracts characters that are erroneously recognized. To do.

Then, the position and the information of the number of characters from the beginning of this area are stored in the memory 14. Also, the correct category of this character is the memory 14
Remember.

(4): Next, in the CPU 11, the memory 14
Image information (original image data) corresponding to the erroneously recognized character is cut out from the optical disc device 18 by using the above-mentioned information regarding the erroneously recognized character and sent to the character recognizing unit 5 to activate the character recognizing unit 5.

(5): In the character recognition unit 5, the image processing unit 21 extracts a feature amount from the cut out image data. Then, in the character recognition unit 5, from the extracted feature quantity and the feature quantity of the category stored in the recognition dictionary 9,
The recognition dictionary is updated so that the dictionary is optimal for recognition.

There are the following methods for updating the recognition dictionary (all of which are conventionally used). : A method of weighted averaging the feature amount of the correct character category of the dictionary and the feature amount of the character that is erroneously recognized.

Clustering is performed on the feature amounts of all categories in the dictionary with the feature amounts of the characters that are erroneously recognized, and the closest category is searched. Then, if the category is not the correct category of the erroneously recognized character, the feature amount of the erroneously recognized character is added to the dictionary of new characters.

The above processing is performed for all the files stored in the optical disk device 18, and the recognition dictionary is learned. §7: Description of document data and description of processing of document structure analysis unit ... See FIG. 3. FIG. 3 is an explanatory diagram of document data. Hereinafter, the processing of the document data (image data and character data) stored in the optical disk device and the document structure analysis unit will be described with reference to FIG.

As described above, a document (printed document) such as a newspaper clipping is input by the scanner 2 and then binarized to be stored in the image memory 3 as image data. The binarization process in this case is as follows.

For example, the histogram of the read densities is taken, and the density is set to be slightly darker than the maximum frequency. Recently, it is common to use a binary algorithm attached to the scanner body.

This is to change the threshold value for each line read by the automatic exposure degree measurement from a variable threshold value set by the user so that the background density and the pattern density are distinguished as much as possible.

The document data input as described above is
As shown in the figure, it is composed of "character area 1", "character area 2", "character area 3", "figure area 1", "photograph area 1" and the like. That is, there are a "character area" in which characters are gathered and whose circumscribed area is a rectangle, a "photo area" in which the circumscribed area of a photograph is surrounded by a rectangle, and a "chart area" in which the circumscribed area of a picture is a rectangle.

By the way, the document structure analysis unit 4 performs the following three processes. : Calculate the coordinates and size of each of the "character area", "photograph area" and "figure area" in the document and store them.

In the area determined to be the “character area”, vertical writing and horizontal writing are further determined from the direction of the character string (row) in the area, which is also stored. : For the text area in the "photo area" or the "chart area", if the closeness between the picture or the graphic area and the text area is smaller than a certain threshold, the description of the picture or the graphic is displayed. It is determined that there is, and the information is also stored.

§8: Description of Each Area Extraction Process of Document Data--See FIG. 4 FIG. 4 is an area extraction / determination process flowchart. Hereinafter, the area extraction / determination processing will be described with reference to FIG.
It should be noted that S1 to S9 represent respective process numbers.

This processing is a document structure in which a document is input and the input image (image data) stored in the image memory 3 is scanned to extract "character area", "figure area", and "photo area". This is the processing of the analysis unit 4.

First, the coordinates of the circumscribed rectangle of the block of black pixels are obtained by tracing the outline of the block (S1). Next, a histogram of the height of the rectangle is obtained, and it is determined that the height is equal to or higher than the threshold value and the peaks of the histogram are interrupted (S2).
Then, only the rectangles within this range are extracted (S
3). In this case, only a small rectangle remains.

When the small rectangles are at a distance equal to or less than the threshold value and are combined into one to form a large rectangle, two small rectangles are integrated into one rectangle.
This process is repeated to create a large character rectangular area and store it in the memory 14 (S4).

After that, the character rectangular area is removed from the original image (S5), and first of all, from the remaining image, a portion constituted by a line is extracted. For this, for example, the image is scanned with a 3 × 3 line segment detection mask to check a portion that is considered to be a line segment.

In a region where there are many places which are considered to be line segments, a portion which is continuous with black pixels is stored as a "graph region" (S6). After the above processing is completed, the "graph area" is removed from the original image (S7), the remaining image is regarded as a "photographic area", and a circumscribed rectangle of continuous black pixels is obtained and stored (S7).
8). Then, if overlap occurs in each of the above areas,
Perform processing to prioritize "character areas" to eliminate overlap (S
9).

§9: Description of Vertical Writing / Horizontal Writing Judgment Processing of Character Area--See FIG. 5 FIG. 5 is a flowchart of vertical writing / horizontal writing judgment processing of a character area. Hereinafter, the area extraction / determination processing will be described with reference to FIG.

It should be noted that S11 to S14 represent respective process numbers. Further, this process is a process performed by the document structure analysis unit 4. First, the arrangement of the circumscribing rectangles of black pixels representing characters in the character area is investigated (S11). Then, the coordinates of the rectangle are horizontally scanned, assuming horizontal writing.
If the upper and lower coordinates of the upper left point coordinate and the upper right point coordinate are within a certain threshold range by this scanning, it is determined to be horizontal writing (S12).

In the same manner, assuming that the rectangular coordinates are written vertically, the rectangular column is vertically scanned and the left and right coordinates of the upper left point coordinate and the upper right point coordinate are within a certain threshold range. For example, it is determined to be vertical writing (S13).

If either one of the judgments is made, that is adopted. If both judgments are made, the one having the smaller average distance from the adjacent rectangle is selected (S14). §10: Description of Document Data--See FIG. 6 FIG. 6 is an explanatory diagram of document data. The illustrated document data example is an example of the document data stored in the optical disc device 18.

In this example, the document data is "page",
Each of the “photograph area”, “figure area”, and “character area” has a list structure. This is because it is not possible to predict in advance how many each element will be. Since the character of the recognition result is stored in the structure of the figure, it is possible to specify which character in the character area of the original image (input image) is recognized. (Because the character ID is attached).

Each item of the document data example is as follows. : "Struct PAGE * page;" in the "struct BOOK" section is a pointer to the data of the first page, and "struct IMAGE *
“Image;” indicates a pointer to image data.

[0081]: "struct PAG" in the "struct PAGE" section
"E * page;" is a pointer to the data of the next page "," st
"ruct PHOTO * photo;" is a pointer to the data of the first photo area, "struct PICTURE * picture;" is a pointer to the data of the first figure area, and "struct CHAR * char;" is the data of the first character area. Indicates the pointer.

[0082]: "struct PHO" in the "struct PHOTO" section
"TO * photo;" is a pointer to the data in the next photo area ",
“Int x1, y1, x2, y2;” indicates the rectangular coordinates (upper left, lower right) of the photo area, and “int beside;” indicates the character area number of the description.

[0100]: "struct PICTURE"-"struct PICTURE"
"ICTURE * picture;" is a pointer to the data of the next chart area, "int x1, y1, x2, y2;" is the rectangular coordinates of the chart area (upper left, lower right), and "int beside;" is the text of the description Indicates the area number.

: "Struct CHA" in the "struct CHAR" section
"R * char;" is a pointer to the data in the next character area, "in
"t x1, y1, x2, y2;" indicates the rectangular coordinates (upper left, lower right) of the character area, and "struct CHARS * chars;" indicates a pointer to the recognition result character string information.

[0085]: "struct CHA" in the "struct CHARS" section
"RS * chars;" is a pointer to the next recognition result character string information,
“Int x1, y1, x2, y2;” is the rectangular coordinates (upper left, lower right) of the image that forms the character, “int ID;” is the character ID that identifies the character in the character area, and “int code;” Indicates the code (recognition result) of this character.

: "Int x1, y1," in the "struct IMAGE" section
"x2, y2;" is the range of the image, "char image [1000000];"
Indicates the content of image information (binarized). §11: Description of background job 1 ... See FIG. 7. FIG. 7 is an explanatory diagram 1 of a background job, where A is a structural example of a word dictionary, and B is a misrecognized character detection processing flowchart using the word dictionary. is there.

Description of Word Dictionary--See FIG. 7A As shown in FIG. 7A, the word dictionary of the word dictionary / grammar unit 25 is composed of a plurality of word sets. The order of words is the order specified by JIS.

In this case, each word has a specific code (0, etc.)
Separated by. For example, as shown in the figure,
0, subtropical, 0, ... ". : Description of misrecognized character detection process using word dictionary
-Refer to FIG. 7B Hereinafter, the erroneously recognized character detection process using the word dictionary by the sentence analysis processing unit 24 will be described with reference to FIG. 7B. Note that S21 to S26 indicate respective process numbers.

First, from the character string of the recognition result stored in the optical disk device 18, the place where the Chinese characters are continuous is extracted (S21). Then, the extracted character string is compared with the word dictionary (S22).

As a result, if they do not match (S
23), the extracted character string is left as it is (S24), and the above-mentioned processing (S21) is repeated. However, when they match (when the degree of matching is larger than the threshold value) (S23), the word dictionary is searched for a different portion from the extracted character string, and the "character ID" of the character is obtained. . Further, the correct character code for that portion is obtained from the word dictionary (S25).
Next, the “character ID” and the correct answer code are stored in the memory 14 (S26), and the process ends.

When performing the above processing, the sentence analysis processing unit 2
In 4, the parsing process is performed using the grammar of the word dictionary / grammar unit 25. §12: Description of background job 2 ... See FIG. 8 FIG. 8 is an explanatory diagram 2 of a background job, A is a structural example of a recognition dictionary, and B is a recognition dictionary update processing flowchart.

Description of Structure Example of Recognition Dictionary--FIG. 8A
The reference recognition dictionary is a set of "character codes" and "features" as shown in the figure. For example, the character code of the character “A” is “3021”, and the feature amount is 3, 5,
7, 90, 3, ...

Further, the character code of the character "Muta" is "30".
22 ”, and the feature amount is 1, 4, 7, 8, 45
3, ... : Description of the recognition dictionary update process ... See FIG. 8B Hereinafter, the recognition dictionary update process will be described with reference to FIG. 8B. It should be noted that S31 to S35 represent respective process numbers.

This process is a process performed by the character recognition unit 5 on the basis of the information (information regarding misrecognized characters) obtained by the sentence analysis process by the sentence analysis processing unit 24. In the image processing unit 21 of the character recognition unit 5, first, the character ID and the correct answer code stored in the sentence analysis processing unit 24 are retrieved from the memory 14 (S31). Then, a portion corresponding to the character ID is extracted from the image information and a feature amount vector is generated (S3).
2). When the feature amount vector is obtained in this way, this is set as the "update feature amount vector" (S33).

Next, in the recognition dictionary 9, the dictionary feature amount vector of the correct answer code is added with weighting and normalized (S34). Then, the updated feature amount vector is set as a new dictionary vector (S35).

[0096]

As described above, the present invention has the following effects. : Learning from a large amount of accumulated document data and using an optimized recognition dictionary improves character recognition accuracy.

Since the dictionary update process is performed as a background job, the user is not aware of
The recognition dictionary can be optimized even when the user is not using the device.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a device configuration diagram of an embodiment.

FIG. 3 is an explanatory diagram of document data according to the embodiment.

FIG. 4 is a flowchart of a region extraction / determination process in the embodiment.

FIG. 5 is a flowchart of vertical writing / horizontal writing determination processing of a character area in the embodiment.

FIG. 6 is an explanatory diagram of document data according to the embodiment.

FIG. 7 is an explanatory diagram 1 of a background job in the embodiment.

FIG. 8 is an explanatory diagram 2 of a background job according to the embodiment.

FIG. 9 is an explanatory diagram of a conventional technique.

[Explanation of symbols]

 2A image input unit 3 image memory 4 document structure analysis unit 5 character recognition unit 7 file device 9 recognition dictionary 24 sentence analysis processing unit 25 word dictionary / grammar unit

Claims (4)

[Claims] 1. An image input unit (2A) for inputting a document image
And a document structure analysis unit (4) that analyzes the input document image and divides it into regions with different attributes (character region, chart region, photo region), and extracts various types of information. , A character recognition unit (5) for performing character recognition on a character area using a recognition dictionary (9), and a file device (7) for storing document data (image data, character data, etc.) obtained by each unit. In the filing device with OCR, the document data stored in the file device (7),
A sentence analysis process is performed on the character data in the character area to extract erroneously recognized characters, and a sentence analysis processing unit (24) for obtaining the correct character information is provided. Recognition part (5)
The filing apparatus with OCR, wherein the recognition dictionary (9) is updated to enable optimization of the recognition dictionary. 2. The filing device with OCR according to claim 1, further comprising a word dictionary and a word dictionary / grammar section (25) storing grammatical information, and the word dictionary / grammar section (25) word dictionary and grammar. A filing apparatus with OCR, characterized in that the sentence analysis processing section (24) is used to perform a sentence analysis process. 3. The filing device with OCR according to claim 1, wherein when performing the recognition dictionary update processing by the character recognition unit (5), image data corresponding to the erroneously recognized character is generated based on the information about the erroneously recognized character. A filing apparatus with an OCR, which is characterized in that it is cut out to obtain a feature amount, and the recognition dictionary (9) is updated based on the feature amount. 4. The filing apparatus with OCR according to claim 1, wherein the internal processing is multitasking, whereby a text analysis processing by the text analysis processing unit (24) and a recognition dictionary update processing by the character recognition unit (5). Is a filing device with OCR, which can be always executed as a background job regardless of other processing.