JPH0850631A - Character recognition device - Google Patents

Abstract

(57) [Summary] [Configuration] The misidentification determination unit 103 includes a character recognition processing unit 102.
Characters with low reliability of recognition result by and highly likely to be misidentified are identified. The erroneous recognition area determination unit 104 erroneously recognizes a character that is determined to have a high possibility of being erroneously recognized, a range including each n characters before and after the character, or a range of a line in which more than half of the characters are likely to be erroneous. Judge as an area. The false positive area extraction unit 105 writes an image and a candidate character group in the recognition result file 108 in addition to the recognition result for the characters in the false positive area. [Effect] The size of the recognition result file can be reduced, and there is no inconvenience even when the memory capacity and file capacity of the character recognition device are small. When the erroneously recognized character is corrected using the data of the recognition result file, the character image and the candidate character group necessary for the correction can be displayed, so that the effective correction work can be performed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device.

[0002]

2. Description of the Related Art Heretofore, with respect to character recognition and correction of erroneously recognized characters, it has generally been common to perform character recognition and erroneously recognized character correction successively for each document on the same apparatus. When correcting this erroneous character, a method is often employed in which the original image or the character image is displayed on the display to match the original with the recognition result on the display without shifting the viewpoint to the original (for example, Kaihei 4-34671). Further, in order to enable efficient correction, various methods have been devised, such as displaying a group of candidate characters on a display so that the correct character can be selected from them.

Further, since character recognition and correction of erroneously recognized characters are generally performed as the same application, it is not necessary or necessary to save a candidate character group or character image of the character recognition result. Even when a character image is saved, it is common to save the entire document or the image of the recognition processing target range.

[0004]

The character recognition device has often been realized as a single device, but the number of forms incorporated into devices such as scanners and copying machines is increasing.

In the case of this built-in form, the memory capacity or file capacity of the device is restricted, or there is no character input means such as a keyboard, or even if it exists, its ability is often weak. In such a device, it is a good idea to perform only character recognition, and to correct the erroneously recognized character as another application on another device, and in many cases, it will be unavoidable.

When the character recognition and the erroneous character correction are separated as described above, in order to enable the efficient correction of the erroneously recognized character as described above, the character recognition device side recognizes the recognition result and the candidate character group. Obviously, it is convenient to save the image of the character area as a file. However, if the memory capacity or file capacity cannot be increased on the character recognition device side, for all characters,
It may be impossible to save the image and the candidate character group, and the environment of the device side that corrects the misidentified character may be burdened.

Further, even in a mode in which character recognition and erroneous character correction are executed as different applications on the same character recognition device, it is convenient to store not only the recognition result but also the candidate character group and the image. If the memory capacity or file capacity of is not large,
It is difficult to store images and candidate characters for all characters.

An object of the present invention is to perform character recognition and erroneous character correction separately in a character recognition device and another device, or in the same character recognition device. In a processing mode in which is performed as a separate application, even if the memory capacity or file capacity of the character recognition device cannot be increased, it is intended to enable efficient erroneous character correction. More specifically, an object of the present invention is to reduce the size of a recognition result file in a character recognition device that performs character recognition to generate a recognition result file without sacrificing the efficiency of the operation of correcting a misidentified character. It is in.

[0009]

The invention according to each of claims 1 to 5 is an invention of a character recognition device, and has the following configurations to achieve the above object.

According to a first aspect of the present invention, a means for performing character recognition for a character image on a character input image, a step of storing the recognition result by the means in a recognition result file, and an image of a character with low reliability of the recognition result. Is stored in the recognition result file.

According to a second aspect of the present invention, means for recognizing a character image on an input image, means for storing the recognition result by the means in a recognition result file, characters with low reliability of the recognition result, and the preceding and following characters. Means for storing an image of each n characters (where n is a positive integer) in the recognition result file.

According to a third aspect of the present invention, in the configuration of the first or second aspect of the invention, for a line in which characters with low reliability in the recognition result occupy a predetermined proportion or more, an image of all characters in the line is stored in the recognition result file. It is characterized by being saved.

The invention of claim 4 is characterized in that, in addition to the structure of the invention of claim 1 or 2, it has means for storing a candidate character group for a character whose image is stored in the recognition result file in the recognition result file. And

According to a fifth aspect of the present invention, in addition to the configuration of the third aspect of the invention, there is provided means for storing a candidate character group for a character whose image is stored in the recognition result file in the recognition result file. To do.

[0015]

[Function] It is expected that a character whose recognition result has low reliability has a high possibility of being erroneously recognized and needs to be corrected. According to the invention of claim 1, since the image of such a character is also stored in the recognition result file together with the recognition result, the data of the recognition result file is used as an application different from the character recognition on the same character recognition device. By displaying the image of the characters that need to be corrected when correcting the erroneous characters or when correcting the erroneous characters on another device, you can efficiently refer to the manuscript without referring to the original. You can check the recognition result and make necessary corrections. Moreover, the image is saved only for the characters with low reliability in the recognition result, so the size of the recognition result file can be significantly reduced compared to the case where the image of all characters is unconditionally saved. it can.

In some cases, it may not be possible to accurately determine misidentification only by the reliability of the recognition result in character units. Characters before and after a character with low reliability often have misidentification even if the character has high reliability. In particular, such a case often occurs when a character recognition processing method that does not perform backtracking in character extraction or language processing is adopted. The invention of claim 2 focuses on such a point. According to the invention, the storage range of the image is expanded to a character within a certain range before and after the character with low reliability of the recognition result. Therefore, the size of the recognition result file increases by an amount corresponding to the expansion of the image storage range as compared with the case of the invention of claim 1, but the size of the recognition result file is still sufficiently small as compared with the case of saving the image of all characters. Become. In addition, when correcting erroneous characters, not only the characters with low reliability in the recognition result but also the image of a certain range of characters before and after that can be displayed, so even if the characters before and after are erroneously recognized, they can be accurately identified. And efficient correction work becomes possible.

The recognition rate of some lines in a document may be extremely poor. For example, it is a line such as a heading or an annotation in which characters with the number of points out of the specification are used. When correcting the recognition result of such a line, it is desirable to be able to confirm the character image of the entire line. According to the third aspect of the invention, when the reliability of the character of a certain line or more is low, the storage range of the character image is expanded to the entire line, so that the recognition rate is extremely bad. It becomes possible to easily and accurately correct a misidentified character in a line such as a heading or an annotation.

According to the invention of claim 4 or 5, not only an image but also a candidate character group is stored in the recognition result file for a character that is expected to be misidentified. Therefore, on the same character recognition device or on another device, when you use the data in the recognition result file to correct erroneously recognized characters, you can display the character image and refer to it, or display a candidate character group from which Since correct characters can be selected, efficient correction work is possible. Moreover, the image and the candidate character group (character data) are saved only for the characters that are likely to be misidentified, so the size of the recognition result file can be made sufficiently smaller than when saving it for all characters. .

[0019]

1 is a schematic block diagram showing an example of a character recognition apparatus according to the present invention. Image data such as a document is input from the image input unit 101. This input image data is stored in the image memory 107. In the character recognition processing unit 102, a character image is cut out from the image data stored in the image memory 107 and character recognition is performed. Since this character recognition processing may be performed by the same method as the conventional one, detailed description thereof will be omitted.

The character recognition processing unit 102 determines, for each character, the character code of the recognition result (first candidate character), the character code of the candidate character group of the second candidate and below, the coordinate value of the character image (for example, the circumscribed character image). The coordinate values of the upper left position and the lower right position of the rectangle) are stored in the recognition result information storage unit 106 (in the form of an array for each character in this example). The recognition result information storage unit 1
The information stored in 06 includes, in addition to the above-described information, the reliability of the recognition result or the misrecognition flag, and the area cutout flag.

The misrecognized character determination unit 103 determines the validity of the recognition result (first candidate character) by the character recognition processing unit 102 or the degree of recognition error (reliability). There are various conventional techniques for this determination, but the method of the present invention does not matter. The misidentified character determination unit 103 outputs a result of the determination, and the content of the output is, for example, a numerical value of reliability or a misrecognition-specific flag indicating that misrecognition is expected to be lower than a certain threshold. The output content is stored in the array of the corresponding characters in the recognition result information storage unit 106. Note that the simplest example of the reliability is the reciprocal of the distance between the recognition result and the dictionary.

Next, the erroneous area determining process by the erroneous area determining unit will be described. FIG. 2 is a flowchart showing the overall flow of the processing. The processing unit of this processing is one line separated by a line feed code. FIG. 3 is a flowchart showing details of the processing step 203 in FIG.

In the erroneous recognition area determination unit 104, the recognition result for one line up to the line feed code and its incidental information are read from the recognition result information storage unit 106 and stored in an internal buffer (step 201). The process of selecting the erroneous region (step 203) is performed for one line in this buffer, the details of which are as shown in FIG.

In step 301, the number (N) of characters with low reliability is counted by referring to the reliability or the misrecognition flag in the buffer. When the reliability is a numerical value, characters whose reliability is lower than a certain threshold value are counted as being expected to be misidentified. When the reliability is displayed in the form of a flag for each misrecognition, a flag value indicating that misidentification is expected is counted. Also, the number of characters (M) in the target line is counted.

Next, in step 302, the misidentification rate of the target line = N / M is calculated. Then, in step 303, it is determined whether the false positive rate is 50% or more.

When the false recognition rate is 50% or more, it is determined that the reliability of the recognition result of the entire target line is low. Therefore, in step 304, the area cut-out flag in the recognition result information storage unit 106 for all the characters in the target line is set to "1" (ON). Such an example is shown in FIG.
"X" in FIG. 4 means that the reliability is low.

If the false positive rate is less than 50%, step 30
In 5, the region cut-out flag in the recognition result information storage unit 106 for the character with low reliability and each character before and after the character is set to "1" (ON). The area cutout flags for the other characters are set to "0" (OFF). Such an example is shown in FIG. "X" means low reliability.

In general, the character with low reliability of the recognition result and the area cut-out flag of each n characters before and after the character are set to "
1 ". N is a positive integer. The example shown here is n = 1.
However, it is usually most efficient to set this way. Further, when n = 0 is selected, only the region cutout flag of the character having low reliability is set to "1".

When the above processing is repeated up to the final line, it is judged that the process is finished in step 201, and the next false recognition area extraction unit 10
The processing shifts to the false recognition area extraction processing by 5. In this process, the recognition result file 108 refers to the area cut-out flag in the recognition result information storage unit 106, and recognizes the recognition result and the incidental information and the image useful for correcting the misrecognized character.
Export to. FIG. 6 is a flowchart showing the flow of this false recognition area extraction processing, and the processing contents of each step will be described below.

In step 402, the area cutting flag of the target character in the recognition result information storage unit 106 is checked. When the area cutting flag is "1" (ON),
The character code of the recognition result (first candidate character) of the target character read out from the recognition result information storage unit 106 by cutting out the image of the target character from the image memory 107 using the coordinate values of the image data in the recognition result information storage unit 106. And reliability, a candidate character group flag (set to ON in this case), an image flag (set to ON in this case), a character code of a second or lower candidate character group, an extracted image, an image and a candidate character group The number of variable length bytes in which the number of bytes of the data is set is written in the recognition result file 108 (steps 403 and 404).

On the other hand, the area cutting flag of the target character is "
If 0 "(OFF), the character code and reliability of the recognition result, the candidate character group flag (set to OFF in this case),
Image flag (set to OFF in this case), variable length byte count (set to zero in this case) Recognition result file 1
(Step 404).

Such processing is repeated for each character,
When the processing end of the last character is judged in step 401,
A series of processing is completed.

By the above processing, when the reliability of the recognition result is low and the character is expected to be corrected, or the character and a character within a certain range around the character, or the reliability of the recognition result as a whole is low. The image data and the candidate character group are stored in the recognition result file 108 only in the range of the considered line. Therefore, compared to saving images and candidate character groups for all characters,
The size of the recognition result file 108 can be significantly reduced, and the storage space for the file can be reduced. This is extremely advantageous when the character recognition device (character recognition function) is incorporated in a device such as a scanner or a copying machine having a small memory capacity.

Regarding the characters that are suspected of being misidentified,
Since the image and the candidate character group are stored, they can be easily corrected on a device such as a personal computer other than the character recognition device, and the character recognition device does not need to have a function for correcting a misidentified character. This means that devices that do not have a character input means such as a keyboard or have a poor character input function,
For example, it has an effect of facilitating the incorporation of the character recognition device (character recognition function) into a device such as a scanner or a copying machine.

Using the data of the recognition result file 108 described above, the recognition result can be corrected on a device such as a personal computer. As a matter of course, on the character recognition device, it is also possible to have a configuration for correcting the erroneously recognized character as an application different from the character recognition.
As described above, even in the case of this configuration, the small size of the recognition result file 108 is advantageous in terms of memory capacity or file capacity.

Here, an example of performing the erroneously recognized character correction process as one of applications on a general personal computer will be briefly described. The application program for correcting the misidentified characters reads the data of the recognition result file 108, and first displays the character code of the recognition result on the display, for example, for one line. At this time, the display color and brightness of the character are changed or the display is reversed depending on whether the reliability is high or low. Next, the application program displays the image of the character whose image flag is ON in the line in association with the character on the display. With this, the operator can easily identify the suspicious character and can easily recognize the correct character by referring to the image.

When the operator designates a character to be corrected by a mouse operation or the like, the application program checks the candidate character group flag for the designated character, and if it is ON, displays the corresponding candidate character group on the display. When the correct character is found in the displayed candidate character group, the operator selects the character by operating the mouse or designating the number from the keyboard.
If the correct answer character is not found, the operator inputs the correct answer character using the keyboard. The application program updates the display on the display and corrects the recognition result data on the memory according to the correct character selected or input.

When the characters with low reliability occupy the majority of the characters in the line, the image of all the characters in the line and the candidate character group are stored in the recognition result file 108. Such a case can be seen in a line (for example, a line such as a headline or a comment) in which a character having a number of points out of the specification is used. In this case, since the character image of the entire line is displayed at the time of correction, it is easy to determine the correct character. In addition, since the candidate character group can be displayed, selection and input of the correct answer character becomes easy.

[0039]

According to the invention of claim 1, 2 or 3,
The character image is saved in the recognition result file together with the recognition result only to the extent that correction is expected to be necessary.Therefore, using the data of this recognition result file, different from character recognition on the same character recognition device. By displaying the image of the characters that need to be corrected when correcting the erroneous characters as an application or when correcting the erroneous characters on another device, the efficiency can be improved without referring to the manuscript. The recognition result can be confirmed and necessary correction can be performed. According to the invention of claim 2 or 3,
Although the size of the recognition result file is slightly larger than that according to the invention of claim 1, since the image of the misrecognized character that cannot be judged from the reliability of the recognition result of each character is also stored, those characters are also imaged. Is displayed to enable efficient correction.

Moreover, since the images are saved only for the characters in the range in which correction is expected to be necessary, the recognition result file is not compared to the case where the images of all the characters are unconditionally saved. Can significantly reduce the size of. Therefore, it is possible to save a recognition result file including an image even in a character recognition device that does not have a large memory capacity or file capacity, such as a character recognition device incorporated in a device such as a scanner or a copying machine. Even in a character recognition device that does not have a large memory capacity or file capacity, it is possible to efficiently perform erroneous character correction separately from character recognition.

According to the invention of claim 4 or 5, since the candidate character group for the character in which the character image is stored is also stored in the recognition result file, the character recognition device corrects the misidentified character as a separate application from the character recognition. When you perform the error correction or correct the erroneous characters on other devices, you can display the candidate character group together with the image saved in the recognition result file and select the correct character from among them. Correction work becomes possible. Moreover, since the range of characters in which the candidate character group (character data) is stored is limited, it is possible to suppress an increase in the size of the recognition result file, and thus the benefits described with respect to the inventions of claims 1 to 3 are impaired. Nothing.

[Brief description of drawings]

FIG. 1 is a block diagram showing an example of a character recognition device according to the present invention.

FIG. 2 is a flowchart showing an overall flow of processing of a false positive area determination unit.

FIG. 3 is a flowchart showing details of some processing steps in FIG.

FIG. 4 is a diagram illustrating an example of a processing result by a false positive area determination unit.

FIG. 5 is a diagram showing another example of the processing result by the false positive area determination unit.

FIG. 6 is a flowchart showing a processing flow of a false positive area extracting unit.

[Explanation of symbols]

 Reference Signs List 101 image input unit 102 character recognition processing unit 103 misidentified character determination unit 104 misidentified region determination unit 105 misidentified region extraction unit 106 recognition result information storage unit 107 image memory 108 recognition result file

Claims (5)

[Claims] 1. A means for performing character recognition on a character image on an input image, a means for storing a recognition result by the means in a recognition result file, and an image of a character with low reliability of the recognition result in the recognition result file. A character recognition device comprising: a storing unit. 2. A means for performing character recognition for a character image on an input image, a means for storing a recognition result by the means in a recognition result file, a character having a low reliability of the recognition result and n characters before and after the character (however). a character recognition device for storing an image of n being a positive integer) in the recognition result file. 3. For a line in which characters with low reliability of recognition result occupy a predetermined proportion or more, an image of all characters in the line is stored in a recognition result file. Character recognizer. 4. The character recognition device according to claim 1, further comprising means for storing, in the recognition result file, a candidate character group for a character whose image is stored in the recognition result file. 5. The character recognition device according to claim 3, further comprising means for storing a candidate character group for a character whose image is stored in the recognition result file in the recognition result file.