JPH08287191A - Optical character reader - Google Patents

Abstract

PURPOSE: To provide an optical character recognition device which can enables a user himself or herself to register a character pattern corresponding to the use environment in a recognition dictionary. CONSTITUTION: The user generates a sample document into which characters to be learnt are entered, stores an FC information storage part 4 with correct answer information generated by giving correction answer character codes to the sample document through input operation, and then makes this device to read the sample document. A control part 8 after obtaining a recognition result from the sample document registers a learnt character pattern in the recognition dictionary in a dictionary storage part 2 on the basis of the correct answer information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical character reading device, and more particularly to an optical character reading device having a customizing function that allows a user to register a character pattern according to a use environment in a recognition dictionary.

[0002]

2. Description of the Related Art Since an optical character reader has conventionally used a recognition dictionary in which standard character patterns are registered, the font of a font having a low recognition rate may occur on the user side. In this case, the manufacturer had to maintain the dictionary for each user. In this maintenance work, it was necessary to select character patterns to be added, determine the number of additional character patterns, adjust competition between character categories, etc., and required specialized technology.

[0003]

However, although there is a demand for the user to create a recognition dictionary, it has been difficult for the conventional device to create a highly accurate dictionary on the user side.

Therefore, the present invention has been made in view of the above circumstances, and an object of the present invention is to provide an optical character reading device in which a user can register a character pattern according to a use environment in a recognition dictionary. It is a thing.

[0005]

Means for Solving the Problems The means of the present invention for solving the above problems will be described.

The optical character reader according to the first aspect of the present invention is an optical character reader that registers a learned character pattern of a target in a recognition dictionary by reading a sample document in which a character to be learned is written. A scanner unit that detects an image of a document, a dictionary storage unit that stores a recognition dictionary in which standard character patterns are registered in advance corresponding to character codes, and a character from the image of the document detected by the scanner unit. A recognition unit that cuts out a pattern and outputs a recognition result by collating with the recognition dictionary, and a correct answer information storage that stores correct answer information in which a correct answer character code is given to each learning target character in the sample document based on an input operation. Department,
After controlling the scanner unit and the recognition unit to obtain a recognition result for the sample document, a control unit for registering a learning character pattern in the recognition dictionary based on the correct answer information is provided. .

An optical character reading device according to a second aspect of the present invention includes a display unit, and the control unit registers the learned character pattern and then controls the recognition unit to recognize the learned character pattern. After obtaining the recognition result, the recognition rate of the learning character pattern in the recognition result is obtained for each subset and displayed on the display unit, and the learning character pattern of the designated subset is registered in the recognition dictionary in duplicate. Is.

An optical character reader according to a third aspect of the present invention comprises a display unit, and the control unit controls the recognition unit after registering a learned character pattern and controls the standard character pattern and the learned character pattern. After recognizing and getting the recognition result,
The recognition rate of the character pattern in the recognition result is obtained for each standard character pattern and learning character pattern for each subset and displayed on the display unit, and the standard character pattern or learning character pattern of the designated subset is duplicated in the recognition dictionary. And then register.

An optical character reading apparatus according to a fourth aspect of the present invention includes a display unit, and the control unit controls the recognition unit after registering the learned character pattern and controls the standard character pattern and the learned character pattern. After recognizing and getting the recognition result,
The recognition rate of the character pattern in the recognition result is obtained for each category, displayed on the display unit, and the character pattern of the designated category is registered in the recognition dictionary in duplicate.

An optical character reading device according to a fifth aspect of the present invention includes a display unit, and the control unit displays the learning character pattern on the display unit before registering the learning character pattern and specifies the learning character pattern. Learning character patterns other than the learned character pattern are registered in the recognition dictionary.

[0011]

The operation of the present invention having the above construction will be described.

According to the apparatus of the invention as defined in claim 1,
The user creates a sample document in which characters to be learned are entered, stores the correct answer information in which the correct answer character code is given to the sample document in the correct answer information storage unit, and then reads the sample document into this device. Let After obtaining the recognition result for the sample document, the control unit registers the learned character pattern in the recognition dictionary based on the correct answer information.

According to the device of the invention described in claim 2,
After registering the learning character pattern, the control unit obtains and displays the recognition rate of the learning character pattern for each subset. Even if the learning character pattern is registered, the recognition rate may not reach the target. In this case, the user specifies a subset whose recognition rate does not reach the target. The control unit redundantly registers the learned character patterns of the designated subset in the recognition dictionary. By registering the same character pattern redundantly, the recognition rate is improved.

According to the apparatus of the invention described in claim 3,
After registering the learned character pattern, the control unit obtains and displays the recognition rate for each of the standard character pattern and the learned character pattern for each subset. Even if the learning character pattern is registered, the recognition rate may not reach the target, or if the learning character pattern is registered, the recognition rate of the standard character pattern may be lower than the target. In this case, the user specifies a character pattern of the subset whose recognition rate does not reach the target. The control unit redundantly registers the character pattern of the designated subset in the recognition dictionary.

According to the apparatus of the invention as defined in claim 4,
After registering the learned character pattern, the control unit obtains and displays the recognition rates of the standard character pattern and the learned character pattern for each category. The user specifies a character pattern of a category whose recognition rate does not reach the target. The control unit redundantly registers the character pattern of the designated category in the recognition dictionary.

According to the device of the invention described in claim 5,
The control unit displays the learned character pattern before registering the learned character pattern. The user specifies a learned learning character pattern that is defective. The control unit registers learning character patterns other than the designated learning character pattern.

[0017]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 shows a first embodiment of the optical character reader of the present invention.
It is a block diagram shown in the Example of.

As shown in FIG. 1, the apparatus of this embodiment has a scanner unit 1 for detecting an image of a form, a dictionary storage unit 2 for storing a recognition dictionary, and a character from the image of the form detected by the scanner unit 1. A recognition unit 3 that cuts out a pattern and outputs a recognition result by collating with a recognition dictionary, an FC information storage unit (correct answer information storage unit) 4 that stores various format control (FC) information, and a learning object and the like are stored. Learning object storage unit 5 such as a hard disk
And CRT display, LED (liquid crystal display)
And the like, a input unit 7 including a keyboard, a mouse, and the like, and a control unit 8 for controlling each unit of the apparatus.

Next, the details of each of the above parts will be described.

The scanner unit 1 is equipped with a light source for irradiating light onto a form and a photoelectric converter for receiving reflected light from the form and converting it into an electric signal according to the amount of light, and the entire form is optically optically. The image of the entire form is detected by scanning.

The dictionary storage unit 2 stores a recognition dictionary in which standard character patterns are registered in advance for each character type (hereinafter referred to as "subset") such as alphabetic characters, numbers, and kana characters in correspondence with character codes. ing. Further, the dictionary storage unit 2 can register one character code regardless of whether or not a plurality of character patterns are the same as the standard character pattern. By registering a plurality of identical character patterns for one character code, the effect of improving the recognition rate of the character patterns can be obtained. The “recognition rate” refers to the ratio of the number of correct reading patterns to the total number of patterns to be recognized.

The recognition unit 3 is adapted to perform character recognition processing using, for example, the composite similarity method. In this composite similarity method, the recognition unit 3 cuts out a character pattern from the image of a form detected by the scanner unit 1, compares the character pattern with the recognition dictionary in the dictionary storage unit 2, and calculates a similarity value. Then, the similarity value is converted into a score, and a plurality of candidate characters from the first candidate character having the largest score to the nth candidate character are sequentially selected and output as a recognition result (character code). . Further, when there is a high possibility that the recognition result will be misread, such as when the first candidate character and the second candidate character have the same score, the recognition unit 3 determines that the first candidate character is the recognition result. Rejected characters (for example, "?") Are output instead of the characters.

In addition to the above character recognition processing, the recognition section 3 also generates a learning object relating to a character font designated as a learning target and stores it in the learning object storage section 5. This "learning object"
Includes a character code (including a candidate character code) as a recognition result, a character pattern to be learned (hereinafter referred to as “learning character pattern”), and attribute information, as well as a subset used by the recognition unit 3 in the recognition dictionary. Information and the like that are generated in the process of recognition processing and that are necessary and sufficient for newly generating a recognition dictionary.

The FC information storage unit 4 stores various FC information such as FC information for sample forms and FC information for adding teacher code (correct character information). The FC information for the sample form includes the position information of the reading field necessary for reading the characters on the sample form. The “sample form” is one in which the characters of the font to be learned are printed or handwritten in a fixed order in each reading field. The FC information for adding the teacher code includes information for adding the correct character code (hereinafter, referred to as “teacher code”) to the learned character pattern of each digit of the reading field in the learning object.
The learning target character is not limited to the font, and may be a handwritten character such as a habit character.

The control unit 8 controls the scanner unit 1 and the recognition unit 3 based on the FC information for the sample form to obtain a recognition result for the sample form, and then learns based on the FC information for adding the teacher code. A character pattern is registered in the recognition dictionary. After the learned character pattern is registered, the recognition unit 3 is controlled to recognize the standard character pattern and the learned character pattern to obtain a recognition result, and then the characters in the recognition result are recognized. The recognition rate of the pattern is obtained for each standard character pattern and learning character pattern for each subset, and these are displayed on the display unit 6, and the standard character pattern or learning character pattern of the designated subset is registered in the recognition dictionary in duplicate. Has become. It should be noted that the recognition rate here means the ratio of the number of correct reading patterns to the number of character patterns of "0" to "9" that belong to the subset, in the case of a subset of numbers.

The control unit 8 may obtain the recognition rate only for the learned character pattern and display it for each subset, and register the learned character pattern of the designated subset in the recognition dictionary in an overlapping manner.

Next, the operation of this embodiment will be described with reference to the flow chart shown in FIG. 4 with reference to FIGS. FIG. 2 is a diagram showing an example of a sample form, and FIG. 3 is a diagram showing an example of a simulation result display screen. In the following description of the operation, a case will be described in which a character pattern of a font other than the standard character pattern registered in advance in the recognition dictionary is newly registered in the recognition dictionary.

First, the user creates a sample form 10 in which the characters of the font to be learned are entered in a fixed order, for example, as shown in FIG. 2 (S1). FIG. 2 shows a case in which the characters 101 of the font to be learned are printed in each reading field 100 in numerical order for a subset of numbers or in alphabetical order for a subset of English letters.

Next, the user creates a definition sheet for generating FC information for sample form and FC information for adding teacher code (S2).

Next, the user gives a reading instruction to the apparatus to read the created definition sheet (S3). The control unit 8 controls the scanner unit 1 and the recognition unit 3 based on the user's reading instruction to execute the reading process for the definition sheet. The scanner unit 1 detects the image of the definition sheet, and the recognition unit 3 cuts out a character pattern from the detected image, collates it with a recognition dictionary, and outputs the recognition result to the control unit 8. The control unit 8 generates FC information for the sample form and FC information for adding the teacher code from the recognition result, and stores it in the FC information storage unit 4.

Next, the user gives a reading instruction to the apparatus to read the sample form 10 created in step S1 (S4). The control unit 8 controls the scanner unit 1 and the recognition unit 3 based on the FC information for the sample form to execute the reading process for the sample form 10. The scanner unit 1 detects an image of the sample form 10.
The recognition unit 3 cuts out a character pattern from the image of the reading field 100 designated by the FC information for the sample form and collates it with a recognition dictionary to obtain a recognition result.

Subsequently, the recognition unit 3 generates a recognition result (character code, candidate character code, character pattern, attribute information) in the process of recognition processing of subset information used by the recognition unit 3 in the recognition dictionary. Then, the learning object is generated by adding necessary and sufficient information for newly generating the recognition dictionary, and is stored in the learning object storage unit 5 (S5).

The control section 8 refers to the FC information for adding the teacher code in the FC information storage section 4 to add the teacher code to the learning object (S6).

The control unit 8 takes out the learning object from the learning object storage unit 5, and uses F for adding the teacher code.
The learning character pattern is registered in the recognition dictionary while determining which category in the recognition dictionary the teacher code corresponds to based on the C information (S7).

After the registration in the dictionary is completed, the control unit 8 performs a recognition simulation to check the validity of the dictionary whether the dictionary is correctly created (S).
8). The control unit 8 recognizes the standard character pattern and the learned character pattern using the recognition dictionary in which the learned character pattern is registered, and obtains the recognition rate for each standard character pattern and the learned character pattern for each subset. The control unit 8 may recognize the learning character pattern included in the learning object.

Next, the control unit 8 displays a simulation result display screen 11 as shown in FIG. 3 on the display unit 6 based on the obtained recognition rate (S9). Figure 3 shows handwritten English letters, handwritten Kana characters, handwritten numbers and JIS-B,
It shows a case where the recognition rate 113 is displayed for each standard character pattern and learning character pattern for each subset such as IBM407 and JIS-K. At this time, the recognition rate is displayed in a warning color such as red for the character patterns of the subset that does not reach the designated target recognition rate. Further, on this screen 11, a "learning" key 110 for executing learning, a "learning return" key 111 for returning to the previous learning result, and an "end" key 112 for ending learning are further displayed. Has been done. Each key 110, 111, 112 is executed by the control unit 8 when it is selected by the cursor.

The user looks at the simulation result display screen 11 and determines whether or not learning is required (S10).

When all the recognition rates have reached the target recognition rates, the "end" key 112 on the screen 11 is designated by the cursor to end the registration process. Even if the learning character pattern is registered, the recognition rate may not reach the target, or if the learning character pattern is registered, the recognition rate of the standard character pattern may be lower than the target. In this case, the user selects a standard character pattern or a learning character pattern of a subset whose recognition rate does not reach the target as a learning target, and performs “learning”
The key 110 is designated by the cursor.

The control unit 8 registers the character pattern of the selected subset in the recognition dictionary (S7), and again recognizes the standard character pattern and the learned character pattern using the recognition dictionary (S8). The recognition result is calculated, and the simulation result display screen 11 is displayed again on the display unit 6 (S9). If all recognition rates do not reach the target recognition rate (S10),
The process for learning (S7,
Repeat S8 and S9). When all the recognition rates have reached the target recognition rate (S10), the "end" key 112 on the screen 11 is designated by the cursor, and the registration process ends. When the recognition rate of the previous learning result is higher, the user designates the “learn back” key 111 with the cursor. The control unit 8 deletes the previously registered content from the recognition dictionary in order to return to the previous learning result.

According to the first embodiment described above, the user causes the apparatus to read the sample form 10 in which the learning target characters are entered, so that the learning character pattern is registered in the recognition dictionary. It becomes possible for itself to register the character pattern according to the usage environment in the recognition dictionary. Further, since the learned character patterns can be registered in duplicate for each subset, the recognition rate of the entire subset can be improved and the labor for registration for each category can be saved. Furthermore, not only learning character patterns but also standard character patterns can be registered in the recognition dictionary in duplicate,
It is possible to improve the recognition rate of the learning character pattern without lowering the standard character pattern.

Next, a second embodiment of the present invention will be described with reference to FIG. FIG. 5 is a diagram showing an example of a category-based simulation result display screen.

In the first embodiment, the recognition rate is displayed for each subset in step S9, but in the present embodiment, the control unit 8 obtains the recognition rate for each category, and the recognition rate is classified by category as shown in FIG. Simulation result display screen 12
Is displayed on the display unit 6, and the character patterns of the designated category are registered in the recognition dictionary in an overlapping manner.

Note that FIG. 5 shows 0 as a subset of handwritten numbers.
The case where the recognition rates 123 of the categories 9 to 9 are displayed is shown. Further, on the screen 12 of FIG. 5, a “learning” key 120 for executing learning, a “learning return” key 121 for returning to the previous learning result, and an “end” key 122 for ending learning. Is displayed. Each key 1
20, 121 and 122 are executed by the control unit 8 when they are selected by the cursor. On the screen 12 of FIG. 5, as in the first embodiment, the marking 12a may be displayed in a warning color such as red for a character pattern of a category that does not reach the designated target recognition rate.

According to the second embodiment as described above, since the character patterns can be registered redundantly, the recognition rate of the character patterns can be improved individually.

Next, a third embodiment of the present invention will be described with reference to FIGS. 6 and 7. FIG. 6 is a diagram showing an example of a character pattern display screen and an example of a teacher code list display screen.

In the first and second embodiments, the learning character pattern is registered immediately after the teacher code is added in step S6, but in the present embodiment, before the control unit 8 registers the learning character pattern. Further, the character pattern display screen 13 as shown in FIG. 6 is displayed on the display unit 6, and only the learning character patterns other than the designated learning character pattern among the plurality of displayed learning character patterns 130 are registered.

FIG. 6 shows a learning character pattern 130 of a subset of handwritten numbers and a character code 13 which is the recognition result.
1 and the teacher code 132 given to the learning character pattern 130 are displayed. Further, on the screen 13 of FIG. 6, a “deletion” key 133 for deleting the pattern designated by the cursor from the pattern file in the dictionary storage unit 2, and the attribute information of the learned character pattern is a certain pattern. A "copy" key 134 for copying to, a "search" key 135 for searching a pattern file for a pattern of a specified condition,
A "sort" key 136 for sorting the contents of the pattern file according to the specified conditions, and a "print" key 1 for printing the contents of the learning object on the connected printer.
37, a "save" key 138 for storing the current state in the learning object storage unit 5, and an "end" key 139 for ending the display of the character pattern are displayed.

On this screen 13, a cursor is used to specify a learning character pattern to be deleted and marking (black circle in the figure) 13a is performed, and a "delete" key 133 is specified to pattern the learning character pattern. Can be deleted from the file. For example, as shown in the figure, when the learned character pattern 130 of the numeral 4 has dust 130 ', such a defective character pattern that should not be learned can be eliminated by the above configuration.

On this screen 13, the teacher code 13
If an error is found in 2, the teacher code of that error is 13
Move the cursor to 2 and use keyboard input or
By selecting from the teacher code list display screen 14 as shown in, the correct teacher code can be input and the teacher code can be corrected. In the screen 14 of FIG. 7, by designating a character (“U” in the figure) with the cursor,
The corresponding teacher code can be entered without directly entering the character code.

According to the third embodiment as described above, since only the designated character pattern can be learned, it is possible to eliminate bad patterns that should not be learned and obtain a higher recognition rate. . Further, the user can also copy the pattern, change the teacher code, mark whether or not to register, edit the pattern, rearrange the pattern, and the like.

In each of the above-described embodiments, the case where the character 101 to be learned is entered on the sample form 10 provided with the reading field 100 has been described.
It is also applicable to sample documents that do not have a reading field. That is, create layout information,
The character to be learned can be read by the device side discriminating the reading region based on the layout information or by automatically discriminating the reading region from the image of the document detected by the scanner unit.

The case where the composite similarity method is adopted as the character recognition processing method of the recognition section 3 has been described. However, instead of the composite similarity method, the contour structure matching method is adopted.
Alternatively, both the composite similarity method and the contour structure matching method, and another method may be adopted, and the method of character recognition processing may be selected according to the type of character and the like so that learning for the learned character pattern can be performed. For example, by adopting the contour structure matching method for handwriting, or by adopting both the composite similarity method and the contour structure matching method, a higher recognition rate is obtained as compared with the case of only the composite similarity method, Higher learning effect can be obtained.

[0054]

According to the present invention described in detail above, the following effects can be obtained.

According to the first aspect of the invention, the user reads the sample document in which the characters to be learned are read by the apparatus, and the learned character pattern is registered in the recognition dictionary. Therefore, the user himself / herself uses the environment. It is possible to provide an optical character reading device capable of registering a character pattern according to the above in a recognition dictionary.

According to the second aspect of the present invention, since the learning character patterns can be registered in duplicate for each subset, the recognition rate of the entire subset can be improved and the labor of registration for each category can be omitted. be able to.

According to the third aspect of the present invention, not only the learned character pattern but also the standard character pattern can be registered in the recognition dictionary in an overlapping manner. Therefore, the recognition rate of the learned character pattern can be maintained without lowering the standard character pattern. It becomes possible to improve.

According to the fourth aspect of the invention, since it is possible to register the character patterns in duplicate, it is possible to improve the recognition rate of the character patterns individually.

According to the fifth aspect of the invention, it is possible to learn only the designated character pattern.
It becomes possible to obtain a higher recognition rate by eliminating bad patterns that should not be learned.

[Brief description of drawings]

FIG. 1 is a block diagram of this embodiment.

FIG. 2 is a diagram showing an example of a sample form of the present embodiment.

FIG. 3 is a diagram showing an example of a simulation result display screen of the present embodiment.

FIG. 4 is a flowchart showing the operation of registration processing of the present embodiment.

FIG. 5 is a diagram showing an example of a category-based simulation result display screen according to the present embodiment.

FIG. 6 is a diagram showing an example of a character pattern display screen according to the present embodiment.

FIG. 7 is a diagram showing an example of a teacher code list display screen according to the present embodiment.

[Explanation of symbols]

 1 Scanner Section 2 Dictionary Storage Section 3 Recognition Section 4 FC Information Storage Section (Correct Answer Information Storage Section) 6 Display Section 8 Control Section 10 Sample Report (Sample Document) 11 Simulation Result Display Screen 12 Category Simulation Result Display Screen 13 Character Pattern Display Screen 101 Learning target characters 113, 123 Recognition rate 130 Learning character pattern 131 Recognition result 132 Teacher code (correct character code)

Claims (5)

[Claims] 1. An optical character reading device for registering a learning character pattern of a target in a recognition dictionary by reading a sample document in which characters to be learned are written, and a scanner unit for detecting an image of the document. , A dictionary storage unit that stores a recognition dictionary in which standard character patterns are registered in advance corresponding to character codes, and a character pattern is cut out from an image of a document detected by the scanner unit and collated with the recognition dictionary to obtain a recognition result. A recognition unit that outputs a correct answer information storage unit that stores correct answer information that gives a correct answer character code for each learning target character to the sample document based on an input operation, and controls the scanner unit and the recognition unit. After the recognition result for the sample document is obtained, the learning character pattern is registered in the recognition dictionary based on the correct answer information. Optical character reader to. 2. A display unit, wherein the control unit registers the learning character pattern, controls the recognition unit to recognize the learning character pattern to obtain a recognition result, and then learns the learning character in the recognition result. 2. The optical system according to claim 1, wherein the recognition rate of the pattern is obtained for each subset and displayed on the display unit, and the learned character patterns of the designated subset are registered in the recognition dictionary in an overlapping manner. Character reader. 3. A display unit, wherein the control unit registers the learned character pattern, controls the recognition unit to recognize the standard character pattern and the learned character pattern, obtains a recognition result, and then recognizes the recognition result. The recognition rate of the character pattern in the result is obtained for each standard character pattern and learning character pattern for each subset and displayed on the display unit, and the standard character pattern or learning character pattern of the designated subset is duplicated in the recognition dictionary. The optical character reader according to claim 1, wherein the optical character reader is registered. 4. A display unit, wherein the control unit registers the learned character pattern, controls the recognition unit to recognize the standard character pattern and the learned character pattern, obtains a recognition result, and then recognizes the recognition result. The recognition rate of the character pattern in the result is obtained for each category, displayed on the display unit, and the character pattern of the designated category is registered in the recognition dictionary in duplicate. The optical character reading device described. 5. A display unit is provided, wherein the control unit displays the learned character pattern on the display unit before registering the learned character pattern, and learns a learned character pattern other than the designated learned character pattern in the recognition dictionary. The optical character reader according to claim 1, 2, 3, or 4, wherein the optical character reader is registered in.