JPH0330190B2 - - Google Patents

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to a character recognition device that recognizes characters written on paper or the like with a writing instrument, etc., and determines a character code representing the character. This relates to the recognition of modified characters such as abbreviations that are unique to each person.

[Prior art]

Most conventional character recognition devices mainly recognize characters written by an unspecified number of people, so they are thought to represent the average deformation of character patterns written by many people. Character recognition was performed using one type of recognition dictionary. For this reason, there is a high probability that abbreviations unique to each person or characters that are significantly deformed will be misrecognized or rejected. In order to prevent this, the person filling out the form was requested to follow the standard character form specified by JIS.

However, there was considerable resistance on the part of those filling out the form, which required them to fill in the form in imitation of the standard character form, and this was a cause of hindering the widespread use of character recognition devices.

The method proposed to eliminate this problem is also called the filler-limited method, which is similar to a speaker-limited speech recognition device and uses a recognition dictionary determined for each filler. An example of this method is disclosed in Japanese Patent Publication No. 58-53393, which is equipped with a plurality of recognition dictionaries, and by inputting a filler code, the recognition dictionary for the filler is selected and used. It is something. Although this method can deal with deformation of the character shape depending on the person writing it, it is difficult to say that it is practical because it has the following drawbacks.

That is, the first drawback is that in order to be usable by a large number of fillers, many types of recognition dictionaries are required, and a huge amount of storage capacity is required to store them. One type of recognition dictionary for recognizing kanji requires several megabytes of storage capacity, and when storing 100 types of recognition dictionaries, even if an external storage device such as a magnetic disk device is used, the storage device is extremely expensive. Become something.

Many types of recognition dictionaries are stored in a relatively low-cost external storage device such as a flexible disk device,
1 selected from among them by the filler code
Although it is also possible to transfer only the type recognition dictionary to a high-speed memory for use, this has the disadvantage that the time required for transfer increases.

The second drawback lies in the difficulty of creating a recognition dictionary for each person who fills in the information. Current technology has not yet achieved the ability of the character recognition device itself to learn from character patterns written by a person and create a recognition dictionary for that person. Since human intervention is required, creating many types of recognition dictionaries requires a large amount of time from many engineers.

In addition, when creating a recognition dictionary, it is said that at least several dozen character patterns are required per category (per character code), and for kanji that target more than several thousand categories. It is almost impossible to have the person write all the character patterns necessary to create a recognition dictionary.

A third drawback is that the recognition dictionary cannot be used by a person who has not registered it. For this reason, there was a drawback in that an unregistered filler had to fill in standard character patterns and use a recognition dictionary for standard character patterns.

[Summary of the invention]

This invention was made in order to eliminate the drawbacks of the conventional ones as described above, and this invention provides a recognition dictionary for standard character patterns and a recognition result correction table that can be updated according to the person who wrote it. It is equipped with a cathode ray tube display device (generally a character display means) and a keyboard (generally a character input means) which constitute a man-machine interface between the person writing the data and the device, and this man-machine interface When a user corrects character recognition results through the system, correction information specific to the filler is stored in a recognition result correction table, and corrections are made using the recognition result correction table in which this correction information is stored. It is.

[Embodiments of the invention]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the present invention, in which 1 is a sheet of paper, 2 is a scanning means, 3 is a recognition means, 4 is a recognition dictionary, 5 is a control means, 6 is a storage means, 7 is a character display means, and 8 is a character input means. The recognition dictionary 4 is a recognition dictionary that targets standard character patterns. Therefore, the operations 1, 2, 3, and 4 are similar to those of a conventional device equipped with a recognition dictionary for standard character patterns, and are well known, so a description thereof will be omitted.

The storage means 6 stores a recognition result correction table (described in a later section), the control means 5 is composed of a general-purpose microprocessor, etc., and the recognition means 3
The character code output as the character recognition result is corrected by comparing it with the contents of the recognition result correction table stored in the storage means 6, the character code to be the final recognition result is determined, and the determined character code is input to the character display means 7.

The character display means 7 has a built-in font memory in which the dot pattern of the character is stored using the character code as an address, and reads out the dot pattern of the character from the font memory according to the input character code and displays it on the cathode ray tube display screen. to be displayed.

The filler observes the character arrangement displayed on the character display means 7, and if there is an incorrectly recognized character in the character arrangement, the incorrectly recognized character is changed to the correct character by inputting from the character input means 8. Replace. If the character to be input is a Roman character, etc., the character code of that character can be input by operating the key on the keyboard of the character input means 8 with the character written on it, and the character to be input can be input. If the character is a kanji, the desired kanji can be input by using kana-kanji conversion technology and keying in the kana characters. In the control means 5, there is a character display means 7.
A refresh memory is built-in to store a character code arrangement at a position corresponding to the character arrangement displayed on the screen. When you key in the desired character after setting the cursor position, the character code at the refresh memory location corresponding to the cursor position will be corrected to the character code of the keyed-in character, and the character code will correspond to this corrected character code. A dot pattern of characters is displayed on the character display means 7. When all the incorrectly recognized characters have been corrected, the filler uses the character input means 8 to send a completion signal to the control means 5.

The control means 5 recognizes the contents of the refresh memory at the time of receiving this end signal as correct, and updates the contents of the recognition result correction table in the storage means 6.

FIG. 2 is a diagram showing an example of the recognition results obtained by the operation of the present invention, where 9, 10, 11, and 12 are characters written on paper 1, and 13, 14, 15, and 16 are characters that are written on paper 1.
0, 11, and 12 represent recognized character codes, respectively. In input character 9, the question was written in an abbreviation, so it was misrecognized as "same", and in input character 10, the layer was written in an abbreviation, so it was misrecognized as "乃", and input character 1
In input character 1, ``ki'' was entered as an abbreviation, so it was recognized as ``san'', and in input character 12, ``tip'' did not exist in the recognition dictionary 4 (that is, ``tip'' is a foreign character), so it was misrecognized as ``yaku'' (translation). Here is an example.

FIG. 3 is a diagram showing the initial state of a part of the recognition result correction table stored in the storage device 6. As shown in FIG.
In the figure, 13a is the recognition result, 17a is the final recognition result, and 18a is the frequency. Since this is the initial state, the recognition result and the final recognition result match, the frequency is 1, and the other fields of the final recognition result are empty. The list of recognition results in Figure 3 has been temporarily collected and arranged for the convenience of explanation related to Figure 2, and the arrangement in the recognition result correction table is based on the bit pattern of the character code for easy addressing. They are arranged according to the JIS kanji code.

When a person uses the device shown in FIG. 1 for the first time, he or she inserts the flexible sheet containing the initial recognition result correction table into a storage device such as a flexible disk device, and then inserts the sheet 1.
If input characters 9 to 12 (in the case of Fig. 2) are written by hand, the recognition means 3 will input the characters 9 to 12 (in the case of Fig. 2) by the same operation as in the conventional device.
The recognition results shown in 3 to 16 are output. The control means 5 makes corrections by referring to the recognition result correction table, but in this case, as shown in FIG. 3, since the recognition result 13a and the final recognition result 17a are the same, no correction is made. Figure 2 13-16
The recognition results shown in are displayed as they are on the character display means 7.

As explained above, the person filling in the information uses the character input means 8 to replace the misrecognized characters with correct characters.
By this replacement operation, the character pattern displayed on the character display means 7 is modified, the character code in the refresh memory of the control means 5 corresponding to this character is modified, and the recognition result modification table is updated. The display on the character display means 7 and the modification of the character code in the refresh memory have already been explained, but for example, when "same" displayed on the character display means 7 as the recognition result 13 is modified to "between". The recognition result modification table is modified as follows.

That is, when the recognition results 13 to 16 shown in FIG. The data is temporarily written to the data memory area in 5. FIG. 3 can be seen as the written memory.
However, in FIG. 3, the lines ``interval'', ``layer'', and ``machine'' are written together for reference purposes and do not need to be written in the control means 5. The input from the recognition means 3 is corrected using the part written in the control means 5 from the recognition result correction table, but as shown in FIG. As explained, no correction is made and the character display means 7 shows the following:
For example, it is displayed as "same title" 13. When the filler corrects "same" to "question", control means 5
The character code (same) in the refresh memory is corrected to the character code (question), and at the same time, the first column of the final recognition result in the row with the recognition result "same" in Figure 3 becomes the character code "question". Corrected its frequency 1
is written, and the data previously stored in the first column is shifted to the second column.

When the replacement with the correct character is completed, the person filling in the information sends a completion signal to the control unit 5 using the character input means 8. At that point, part of the recognition result correction table in the control section 5 is as shown in FIG. In Figure 4, "Title" 13c, "High" 14a,
Since "machine" 15c and "another" 16b were not modified, 1 was added to the frequency, making it 2.
In the final recognition result array, those with higher frequencies are placed in the front column, and those with the same frequency are placed in the previous column with character codes that appear later.

When the control section 5 receives the end signal, it corrects the recognition result correction table in the storage means 6 according to the contents corrected in the control section 5.

Further, the recognition means 3 outputs a plurality of character codes for one character pattern and ranks them according to their likelihood, and displays the plurality of character codes as a dot pattern of each character on the character display means 7. However, it can also be configured so that the person filling in the form selects one of the characters.

As described above, the information of the recognition result correction table corresponding to the person who filled in the information is sequentially accumulated in the flexible sheet, but when the person who filled in the information finishes using the device shown in Fig. It is sufficient to take out the flexible sheet to be used from the storage means 6, store it separately, and attach it to the storage means 6 again the next time the apparatus of FIG. 1 is used.

For example, if the contents of the recognition result correction table are
If it is as shown in the figure, input characters 9 to 9 in Figure 2
12 is input again, the correct character of the final recognition result is displayed on the character display means 7, and the person filling in only needs to confirm this and send an end signal.In this case, the recognition result correction table shows the correct character of the final recognition result. is updated. FIG. 5 shows this updated recognition result correction table. In FIGS. 3 to 5, the lines marked with diagonal lines outside the left frame are portions that are not directly related to the input characters shown at 9 to 12 in FIG. 2, as explained with reference to FIG.

FIG. 6 is a diagram showing an example of a part of the contents of a recognition result correction table for a particular person who has used the apparatus of the present invention many times. The person filling this out uses abbreviations as shown in Figure 2 9-12, so the recognition result is "Question" 13a,
"Layer" 14b, "machine" 15a are mostly caused when character patterns of other characters are misrecognized, and "title" 13c, "taka" 14a, "machine" 15c,
“Another” 16b indicates that most of the images were correctly recognized. Also, “No” 14c and “Zan” 15b (Sun)
is rarely used by this author, and is referred to as "layer"
and "machine" to indicate that it has been modified. Also “same” 1
3b is often recognized as "same" 13b from "same", "question", and "ma", and the final recognition result is determined by the frequency of use of these characters. ``Translator'' 16a is also frequently used and is often finally recognized as ``translator'', but this shows the fact that the external character ``tei'' has been recognized as ``translator''.

FIG. 7 is a diagram showing an example of the display of the character display means 7 in FIG. ” and “another” are present in the first column (Fig. 6, 16).
The contents of each line of a and 16b) are shown. There are 2 people who filled in the information.
Since the display of 19 can be corrected by selecting from the display of 0 and 21, it becomes easy to input the character code of the character to be corrected.

As described above, according to the present invention, it is possible to recognize abbreviations unique to the person who filled in the information. For example, as shown in FIG. It may be mistakenly recognized as "machine". However, this person hardly uses ``no'' and ``san'' as shown in Figure 6, so this is not a problem.

The number of character types that each person actually uses as handwritten characters is relatively limited. For example, in a character recognition device that targets kanji, the number of character codes in the recognition dictionary 4 is about 3000, which is the JIS first level.
Even if a 3000-character character code were arranged in the recognition result column of the recognition result correction table, about 800 characters would be used if limited to each person who entered it, and the recognition result correction table would be about 800 characters. The table also reads and writes the 800 or so character types mentioned above. For example, sentences created by electronics engineers almost never contain characters such as ``ノ'' or ``san''. is far more effective.

For example, even if you exclude ``san'' from the recognition dictionary 4, if you write the abbreviation for ``ki'', the probability that this will be recognized as ``ki'' is low, and other characters will be recognized. Since this is often the case, recognition means 3 recognizes it as "san".
It is better to use the control means 5 to correct it to "machine".

Although the use of frequency in character recognition has been done in conventional devices, the use of frequency in this invention is essentially different. For example, in conventional devices, the recognition means outputs the character codes of multiple recognition candidate characters and the degree of similarity of their character patterns, and the most likely character code is determined by a function of this degree of similarity and the frequency of use of the character. If the correct characters are not included in the recognition candidate characters, they will not be recognized correctly.

In addition, in Kana-Kanji conversion, frequency of use is sometimes used as one of the criteria for selecting homophones;
This does not cause a change in the types of characters that can be selected from the same reading, and is different from the recognition result correction table of the present invention, which is used to change the final recognition result.

In addition, in Figure 6, the recognition result is "same" 1.
The frequency of the final recognition result of 3b is "same" 65, "question"
60. Although it is difficult to avoid confusion between the three parties in "between" 55,
This is a limitation of the character recognition device according to the present invention, and is a problem that must be corrected through the man-machine interface.

In the above explanation, it is assumed that the operator who observes the character display means 7 and operates the character input means 8 is the same person who wrote the characters on the form 1. Needless to say, there is no problem even if they are not the same person.

Furthermore, although it has been explained that the recognition result correction tables for each filler are stored in separate flexible sheets, it is possible to store the recognition result correction tables for all fillers in the storage means 6 and to correspond to each filler. By inputting an identification code etc. from the character input means 8, the recognition result correction table corresponding to the person filling in the form is selected, or a character or identification code identifying the filler is written in a predetermined position on the sheet 1. By recognizing this, it is possible to select the recognition result correction table corresponding to the person who filled in the information. The recognition result correction table only needs to store character codes, and can be stored in a relatively small capacity storage device.

If it is not possible to identify the person who filled in the information, the output of the recognition means 3 is sent directly to the character display means 7 via the control means 5.
It should be displayed in .

In addition, in the above explanation, from the recognition means 3, 1
Although only one character code was output in response to one character pattern, it is also possible to output multiple character codes in response to one character pattern along with the degree of similarity of the character code to the reference pattern. The control means 5 can also refer to the degree of similarity for each character code to determine the final recognition result.

Furthermore, it can be applied to online recognition in which characters handwritten by a person writing on a tablet or the like are recognized in real time. Online recognition often uses stroke order information that easily shows the characteristics of each person writing the information, and the effects of this invention are also likely to be exhibited.

〔Effect of the invention〕

As described above, according to the present invention, the final recognition result is determined by the recognition result correction table prepared for each filler, and the recognition result correction table is expanded and revised each time it is used. It is possible to accurately recognize the written characters.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of this invention, FIG. 2 is a diagram showing an example of recognition results obtained by the operation of this invention, and FIGS. FIG. 7 is a diagram showing a part of the contents of the recognition result correction table of the invention, and FIG. 7 is a diagram showing an example of display on the character display means of FIG. 1. DESCRIPTION OF SYMBOLS 1...Paper, 2...Scanning means, 3...Recognition means, 4...Recognition dictionary, 5...Control means, 6...Storage means, 7...Character display means, 8...Character input means.

Claims (1)

[Scope of Claims] 1. A scanning means for photoelectrically converting characters, etc. written on paper, etc. and outputting an electric signal representing a character pattern of the characters, etc., a standard pattern of characters to be recognized, etc., and each standard pattern. a recognition dictionary that stores correspondences with character codes of characters, etc. corresponding to characters, and compares the character pattern output from the scanning means with a reference pattern in the recognition dictionary to match the character pattern output from the scanning means. A recognition means for determining character codes, which arranges all the character codes in the recognition dictionary into a recognition result list, and writes a plurality of character codes and numerical values indicating their frequency of appearance for each character code in the recognition result list. A recognition result correction table is provided for each person who may fill in the above form, etc., and the recognition result correction table is provided with the final recognition result and frequency information of the recognition result correction table in the initial state. Means for writing the character code of the corresponding recognition result RAN in the first position of the frequency RAN and writing the number 1 as the frequency, and selecting and using the recognition result correction table corresponding to the person filling out the form, etc. for each person filling out the form, etc.; Input the character code determined by the recognition means, refer to the above recognition result correction table, and calculate the final recognition result corresponding to the recognition result RAN in which the same character code as the input character code is stored, and the first character of the frequency RAN. A control means for reading out a code, a character display means for converting the character code read by this control means into a dot pattern of the character and displaying it on a display screen, and a character display means for observing the dot pattern of the character displayed on the character display means. and, if necessary, a character input means for inputting the character code of the character to be corrected and correcting the characters displayed on the character display means, after the necessary corrections have been made by this character input means. Write to the recognition result correction table according to the display content of the character display means, and in response to the recognition result, write the character code of the displayed character on the character display means to the final recognition result and the frequency. If the character code is already stored, add the numerical value 1 to the frequency corresponding to the character code, and if the character code of the character displayed on the character display means does not match the frequency with the final recognition result, add the value 1 to the frequency corresponding to the character code. Write the character code and the number 1 as the frequency in the empty space in the result and frequency column, and in the same final recognition result column, place the character code with the highest frequency at the beginning, and the character code with the highest frequency A character recognition device comprising a recognition result correction table augmentation and revision means for arranging a character code that has recently reached a certain frequency among the plural types at the beginning.