JP2953162B2 - Character recognition device - Google Patents

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 for recognizing a character which is conceived by handwritten or printed character image data.

[0002]

2. Description of the Related Art Conventionally, handwritten character image data or character image data printed on a magazine or the like is read as binary image data by an image scanner and compared with a character pattern registered in a dictionary. Character recognition devices have been developed. This device can omit data input from the keyboard,
A wide range of applications can be expected as a very efficient man-machine interface.

However, this character recognition device may erroneously recognize characters due to the habit of handwritten character image data or the blurring or crushing of printed character image data. When the operator finds a misrecognized character from the character string of the recognition result displayed on the screen, there are the following correction methods. First, there is a method in which an erroneously recognized character is instructed with a light pen, and another character is displayed in the order in which the possibility of a correct character is high to correct the character. Next, there is a method of performing correction by displaying a plurality of characters in descending order of the possibility of a correct answer by tapping the keyboard. Further, there is a method of displaying a target character to be recognized and a recognition result character at the same position on the screen, respectively, for correction. According to this method, visual comparison is easy, accurate, and quick.

[0004]

In the conventional character recognition apparatus, when character image data is erroneously recognized at one place, there is a high possibility that the same character image data is erroneously recognized at a plurality of places. . However, even in such a case, the operator had to correct each character one by one. For this reason, the correction work after character recognition takes time, which is a problem. An object of the present invention is to solve such a problem.

[0005]

In order to solve the above-mentioned problems, a character recognition device according to the present invention comprises an input means for inputting a plurality of handwritten or printed character image data, and a character input by the input means. Identification means for comparing image data with a plurality of character patterns registered in a dictionary to recognize a specific character; manual correction means for manually correcting an erroneously recognized character from a group of recognized characters recognized by the identification means; Detects character image data that is identically recognized as a misrecognized character replaced with a corrected character by manual correction means from a group of characters, and converts the pattern of these character image data into a character pattern registered in a dictionary of misrecognized characters. Re-confirmation means for determining that the possibility of misrecognition is high when the pattern is close to the pattern of the image data of the misrecognized character. .

Further, there is provided an automatic correcting means for automatically correcting the character image data determined to have a high possibility of erroneous recognition by the reconfirming means.

[0007]

According to the character recognition device of the present invention, a plurality of handwritten or printed character image data is input by the input means, and the character image data is analyzed by the identification means to determine that the character is a specific character. Be recognized. And
Characters misrecognized by the identification means are manually corrected by the manual correction means. In the reconfirmation means, characters which are recognized as the same as the erroneously recognized characters replaced with the corrected characters by the manual correction means are detected, and the pattern of the image data of these characters is smaller than the character pattern registered in the dictionary of the erroneously recognized characters. It is determined that the possibility of erroneous recognition is high when the pattern is similar to the pattern of the image data of the erroneously recognized character. In addition, the character determined as described above can be automatically corrected by using an automatic correction unit.

[0008]

An embodiment of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a block diagram illustrating the configuration of the present embodiment. As shown in the figure, the character recognition device of the present embodiment includes a CPU 10 for controlling each process, a RAM device 20 for storing a recognition result table, a ROM device 30 for storing processing modules and the like, character input and display, and the like. And an input / output device 40 for performing the operation. The ROM device 30 includes an input module 31 as input means for inputting a plurality of handwritten or printed character image data, an identification module 32 as identification means for recognizing a specific character, and an erroneously recognized character. A manual correction module 33 which is a manual correction means for manually correcting, a reconfirmation module 34 which is a reconfirmation means for rechecking erroneous recognition of another character, and an automatic correction means which is an automatic correction means for automatically correcting erroneously recognized characters. A correction module 35, a correction display module 36 that is a correction display means for highlighting a character determined to have a high possibility of misrecognition, and an input indicating whether a character determined to have a high possibility of misrecognition is misrecognized. A confirmation input module 37 is provided as confirmation input means for causing the user to make a confirmation.
The input / output device 40 includes an image scanner 41 for inputting handwritten or printed character image data,
Display device 42 for displaying character data after recognition
And a keyboard device 43 for correcting erroneously recognized characters, a hard disk device 44 for storing a discrimination dictionary in which a plurality of character patterns are registered as feature amounts, and a switch 45 for use in correction processing. .

Next, the processing contents of this embodiment will be described. The processing of the character recognition device of the present embodiment includes a data reading process by the input module 31, a data recognition process by the identification module 32, and a manual correction module 33.
And an erroneous recognition and correction process by the reconfirmation module 34 and the like.

First, the data input processing will be described.
The data input processing is performed by reading by the operator using the image scanner 41. More specifically, the operator activates the input module 31 under the control of the CPU 10, and presses the image scanner 41 on a character printed on a magazine or the like to read a desired character. By this reading, character image data is input.

Next, the data recognition processing will be described.
In the data recognition process, the operator activates the identification module 32 under the control of the CPU 10, extracts the peripheral features of the input character image data, and stores the feature amount and a plurality of registered characters of the discrimination dictionary stored in the hard disk device 44 in the discrimination dictionary. This is performed by comparing and discriminating the feature amount of the pattern. Here, the extraction of the peripheral feature is performed as follows. First, the circumscribed frame of the character pattern is obtained, and the four sides of the circumscribed frame are divided into four. Then, by counting the area of the divided circumscribed frame and the white area surrounded by the character portion that first meets when viewed from the circumscribed frame, and standardizing this with the entire area,
A feature amount (16-dimensional feature vector) is obtained. For example, as shown in FIG. 2, when the character image area of the character “water” composed of 64 × 64 bits is divided into 16 × 16 dot units, the feature amounts (α ₁ , α ₂ ,..., Α ₁₆ ) , (Α ₁ =
29, α ₂ = 17,..., Α ₁₆ = 15).

The feature amount α _j of the character image data extracted in this way and the feature amount β of a plurality of registered character patterns of the discrimination dictionary stored in the hard disk device 44
The comparison with _ij (i = 1 to n, J = 1 to 16, n is the number of registered characters) is performed as follows.

[0013] Here, i represents the character registered in the dictionary i-th, and D (i) represents the distance from the character registered in the dictionary i-th (the degree of similarity between the two character patterns).
The distance D (i) (i = 1 to n, n
Is the number of registered characters) in ascending order.
In the recognition result table. The recognition result table refers to a table used for erroneous recognition correction processing, and is divided into four types of tables as shown in FIG. 3 according to the type of data stored. These recognition result tables will be described with reference to FIG. FIG. 3A is a conceptual diagram showing the structure of the recognition result table 100. The recognition result table 100 includes character codes of characters registered in the dictionary,
The distance between the characteristic amount of the input character data and the characteristic amount of the character registered in the dictionary and the characteristic amount of the input character data are stored in ascending order of the distance between the characteristic amounts. FIG. 3B is a conceptual diagram illustrating the structure of the recognition result table 110. In the recognition result table 110, a character code, a distance between feature amounts, a feature amount of input character data, and a character image of input character data are stored in ascending order of the distance between feature amounts. The recognition result table 110 is suitable for displaying character data of an input image on the screen of the display device 42. FIG. 3C is a conceptual diagram showing the structure of the recognition result table 120. Recognition result table 1
20 stores a character code, a distance between feature amounts, and a character image of input character data in ascending order of the distance between feature amounts. Although the data amount of the recognition result table 120 can be reduced as much as there is no data of the distance between the feature amounts, there is a drawback that the feature amount must be extracted from the character image of the input character data at the time of matching. FIG. 3D is a conceptual diagram illustrating the recognition result table 130. In the recognition result table 130, the character codes and the distances between the feature amounts are stored in ascending order of the distance between the feature amounts. Further, only when the ratio of the distance between the first and second feature values is smaller than the fixed value γ, the feature value of the input character data is stored together with these data. By using a variable table structure in this way, the data amount can be reduced.

In the data recognition processing, it is determined that the first character in the recognition result table is the same character as the character of the input image data. This processing is performed for all the input image data, and the recognized character string is displayed on the display device 42.

Next, the erroneous recognition and correction processing will be described with reference to FIGS. FIG. 4 is a flowchart showing an outline of the erroneous recognition correction processing. As shown in the figure, the operator activates the manual correction module 33, the reconfirmation module 34, and the like under the control of the CPU 10, and confirms whether all the character strings displayed on the display device 42 are correctly recognized (step 200). Then, when an erroneously recognized character is found (step 220), the cursor is moved to the character displayed on the display device 42 and a correction input is made from the keyboard 43 (step 230).
With this correction input, another character that has already been recognized is rechecked, and all characters that are likely to be erroneously recognized are extracted (step 240). Then, the correction processing of all the extracted characters is performed (step 250), and the processing returns to step 200 after the correction processing is completed. The above process is repeated, and when all characters have been confirmed, the process is terminated (step 210).

Next, referring to FIGS. 5 and 6, a process of extracting a character having a high possibility of erroneous recognition (the process of step 240 in FIG. 4).
Will be described. FIG. 5 is a flowchart showing the extraction processing by the reconfirmation module 34. Note that this extraction processing includes the recognition result tables 100, 110, 130
Is used. As shown in the figure, by the correction input in step 230, all other recognized characters are rechecked, and the same character as the corrected character is searched (step 242). When the same character is detected, the distance D _ist between the characteristic amount of the detected character and the characteristic amount of the corrected character is stored in the recognition result tables 100 and 110.
(Step 24)
3). A value obtained by adding the statistically obtained correction value δ to the distance D _ist calculated in this way is compared with the first-place distance d _{1 of} the detected character (step 244), and the distance d _{1 is obtained.}
If is larger, it is determined that the detected character is likely to be erroneously recognized (step 245). If the distance d ₁ is smaller, it is determined that the detected character is correctly recognized. Then, after the comparison / determination has been performed for all the characters, the extraction processing is terminated (step 241).

FIG. 6 is a flowchart showing the extraction processing by the reconfirmation module 34. Note that the recognition result table 120 is used for this extraction processing. As described above, the difference between the recognition result table 120 and other tables is that the table does not have feature amount data. For this reason, in the flowchart shown in the figure, in step 246, the characteristic amount of the corrected character and the characteristic amount of the detected character are obtained. Other processing is
This is the same as the processing in the flowchart of FIG.

Next, a specific example of the process of extracting a character having a high possibility of erroneous recognition (step 240) will be described with reference to FIG. In the figure, first, image data is input from a magazine or the like using the image scanner 41, and the feature amount of the image data is compared with the feature amount of characters registered in the discrimination dictionary stored in the hard disk device 44. judge. Then, the recognition result is displayed on the display device 42. The operator confirms this screen display and corrects the second digit character "na" to "su". Through this correction process, a search is made to see if any other character has a character recognized as "na". By this search, the fourth digit “Na” is detected. The feature amount of the second digit character “Na” is $ 28, 15, 25,
..., 21, 15} and the feature amount {2 of the fourth digit “Na”
The distance from 9, 16, 26, ..., 20, 16 is D _ist
= 29. In addition, the feature amount of the fourth digit “Na” and the feature amount of the character “Na” registered in the discriminating dictionary, which are $ 30, 16, 3
The distance from 1,..., 21, 21 ° is d ₁ = 53.
Here, assuming that the correction value δ = 10, (d ₁ = 53)>
Since it satisfies the condition of (D _ist = 29) + (δ = 10), it is understood that the possibility of misrecognition of the fourth digit “na” is high and should be corrected to “su”.

Next, the process of correcting the character thus extracted (step 250) will be described with reference to FIGS. In the correction processing, a method of automatically correcting using the automatic correction module 35, a method of correcting one character at a time by the operator's confirmation using the correction display module 36 and the confirmation input module 37, or a selection of automatic correction and correction by the operator. There is a method to correct above.

FIG. 8 is a flowchart showing the correction processing by automatic correction, and FIG. 9 is a display example of the correction processing by automatic correction. 8 and 9, when the operator corrects the second digit “NA” to “SU” in step 230 (screen 30).
1, 302), and at step 240, the fourth digit and the 19th digit “na” are extracted. Then, the extracted two characters "na" are simultaneously corrected to "su" (step 25).
2). In addition, in order to easily identify the corrected character, the character is displayed as a highlighted character such as a reverse character, a character with a different color, or a blinking character (step 253, screen 303).
Then, after correcting all the extracted characters, the process is terminated (step 251).

FIG. 10 is a flowchart showing a correction process for correcting characters one by one by the confirmation of the operator, and FIG. 11 is a display example of a correction process for correcting characters one by one by the confirmation of the operator. From FIG. 10 and FIG.
When the operator corrects the second digit “na” to “su” at 30 (screens 311 and 312), at step 240, the fourth digit and the 19th digit “na” are extracted. And the extracted 4
The cursor is moved over the character "na" in the digit (step 262), and the character at the moved position is highlighted (step 263). Then, a window for confirming the correction is opened (step 264, screen 313), and the operator presses a line feed key (step 265), and
The character "na" in the digit is corrected to "su" (step 26).
6). After this correction, the process returns to step 262, where 19
A correction confirmation window is opened at the position of the character "na" in the digit (screen 314). Then, the 19th digit character “Na” is also corrected to “su” by the input of the line feed key by the operator. As a result of this correction, there are no other characters that may be corrected, so the cursor is returned to the original position and the process is terminated (step 261).

FIG. 12 is a flowchart showing a correction process in which a user interface is provided with a switch 45 capable of selecting automatic correction or correction by an operator. As shown in the figure, in step 240, a character having a high possibility of erroneous recognition is extracted, and the state of the switch 45 is checked (step 2).
72). If the switch 45 is on, processing by automatic correction is performed (step 273). Switch 4
If 5 is off, the operator performs a correction process (step 274).

FIG. 13 is a flowchart showing a correction process by an automatic correction having a malfunction prevention flag. FIG. 14 is a display example of a correction process by an automatic correction having a malfunction prevention flag. From FIG. 13 and FIG.
When the operator corrects the second digit “na” to “su” (screens 321 and 322), the fourth digit and 1
The ninth digit “na” is extracted. Then, it is determined whether or not the flags of the extracted two characters "na" are "ON" (step 282). If the flag is "OFF", the character "na" is corrected to "su" (step 28).
3). In addition, the corrected characters are displayed as highlighted characters such as reverse characters, characters with different colors, and blinking characters so that the characters can be easily identified (step 284, screen 323).
Then, the flag of the corrected character is set to "ON" (step 285). By this processing, the flag of the character “na” at the second, fourth, and 19th digits is set to “ON”,
Next, even if the operator corrects the third digit "su" to "ma", the characters "na" in the second, fourth and 19th digits are corrected to "ma" in step 253. No (Screen 3
04).

In this embodiment, the recognition of the character pattern is performed by extracting the peripheral features by the superposition method. However, the present invention is not limited to this extraction method, but uses a mesh method or a structural analysis method. Is also good.

[0025]

According to the character recognition device of the present invention, the character image data which is recognized as the same as the erroneously recognized character replaced with the corrected character by the manual correcting means is detected. And
If the pattern of the detected character image data is closer to the pattern of the image data of the misrecognized character than the character pattern registered in the dictionary of the misrecognized character, it is determined that the possibility of misrecognition is high. . In addition, the character image data determined in this way can be automatically corrected using an automatic correction unit. Therefore, the work of manual correction after character recognition is completed in a short time, and the burden on the operator is reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of the present embodiment.

FIG. 2 is a conceptual diagram showing an example of extracting a peripheral feature.

FIG. 3 is a conceptual diagram showing a data structure of a recognition result table.

FIG. 4 is a flowchart illustrating an outline of an erroneous recognition correction process.

FIG. 5 is a flowchart showing an extraction process by a reconfirmation module.

FIG. 6 is a flowchart showing an extraction process by a reconfirmation module.

FIG. 7 is a conceptual diagram showing a specific example of an extraction process.

FIG. 8 is a flowchart illustrating correction processing by automatic correction.

FIG. 9 is a diagram showing a display example of correction processing by automatic correction.

FIG. 10 is a flowchart showing a correction process for correcting one character at a time.

FIG. 11 is a diagram illustrating a display example of a correction process for correcting one character at a time.

FIG. 12 is a flowchart illustrating a correction process.

FIG. 13 is a flowchart illustrating a correction process by an automatic correction having a flag.

FIG. 14 is a diagram illustrating a display example of correction processing by automatic correction having a flag.

[Explanation of symbols]

10 CPU, 20 RAM device, 30 ROM device,
31 input module, 32 identification module, 33
Manual correction module, 34 ... reconfirmation module,
35: automatic correction module, 36: correction display module, 37: confirmation input module, 40: input / output device.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/00-9/82

Claims (5)

(57) [Claims] An input unit for inputting a plurality of handwritten or printed character image data, and comparing the character image data input by the input unit with a plurality of character patterns registered in a dictionary to specify a specific character. An identifying means for recognizing; a manual correcting means for manually correcting an erroneously recognized character from the group of recognized characters recognized by the identifying means; and an erroneous character replaced by the manual correcting means from the group of recognized characters. Detecting the character image data that is recognized as the same as the recognized character, the pattern of the character image data is closer to the pattern of the image data of the erroneously recognized character than the character pattern registered in the dictionary of the erroneously recognized character. A character recognizing device comprising: a re-confirmation unit that determines that the possibility of erroneous recognition is high when the character recognition is performed. 2. The data pattern determination by the reconfirmation means is performed by extracting a peripheral feature of each data pattern and comparing the feature amounts to make an approximate determination. Character recognition device. 3. An automatic correction means for automatically correcting character image data determined to have a high possibility of erroneous recognition by said reconfirmation means.
Alternatively, the character recognition device according to claim 2. 4. The apparatus according to claim 1, further comprising correction display means for highlighting character image data determined to have a high possibility of erroneous recognition by said reconfirmation means. The character recognition device described in Crab. 5. The apparatus according to claim 1, further comprising a confirmation input unit for inputting whether the character image data determined to have a high possibility of erroneous recognition by the reconfirmation unit is erroneous recognition. The character recognition device according to claim 4.