JP3666909B2 - Character recognition apparatus and method - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to a character recognition apparatus and method, and more particularly to a character recognition apparatus and method for determining an optimum candidate character in a recognized character candidate group.
[0002]
[Prior art]
Conventional character recognition will be described. When a character string is input using an input device (for example, a scanner) or an input pen as shown in the flowchart of FIG. 3, the character recognition dictionary is searched to determine whether the character string has been specified. . Whether or not it has been specified is determined, for example, by calculating the distance between the feature vector of each character image and the standard feature vector in the recognition dictionary and comparing the calculated result with a predetermined value.
[0003]
When it is determined that the identification has been made, the recognition result (character code string) is output as it is.
[0004]
Moreover, when it cannot identify, possibility that the recognition result is incorrect becomes high and creates a candidate group. Then, a word is determined by repeatedly searching a separately prepared word dictionary.
[0005]
For example, when “electronic” is input as shown in FIG. 4, the combination of candidate character strings obtained by character recognition is “den thousand”, “electronic”, “electric dr”, “den”, etc. As shown in FIG. 5, a word dictionary is searched for these candidates. As a result, since only “electronic” is stored in the word dictionary, “electronic” is determined and output as a recognition result.
[0006]
[Problems to be solved by the invention]
By the way, in the conventional character recognition described above, if the input character is to be adapted to the case where the input character is a “kanji-kana mixed character string”, it is necessary to store all the kanji-kana mixed words in the word dictionary. For example, in the case of the word “electronic”, all of the four types of words “den”, “electric”, “denko”, and “electronic” must be stored in the word dictionary. Therefore, the capacity of a word dictionary for character recognition mixed with kanji characters is significantly larger than that of a word dictionary with only kanji characters. However, a high recognition rate of characters cannot be expected unless all kanji-kana mixed words are stored in the word dictionary.
[0007]
[Means for Solving the Problems]
And [Action]
The present invention has been made in view of such problems, and it is possible to improve the recognition rate when the character string to be recognized is a word mixed with kanji characters while reducing the size of the dictionary for recognition processing. An object of the present invention is to provide a character recognition apparatus and method.
[0008]
In order to solve this problem, for example, the character recognition device of the present invention has the following configuration. That is,
An input means for inputting a kanji mixed character string;
Character recognition means for recognizing characters of a kanji mixed character string input by the input means, and outputting one or more candidate characters for each character;
Candidate word forming means for forming one or a plurality of candidate words for the character string mixed with kanji by combining candidate characters of each character obtained by character recognition by the character recognition means;
The candidate If word candidate word formed by formation means determines that there are multiple, the readings of the candidate Chinese characters included in the candidate word that has been processed from among the plurality of candidate words to one or more retrieving process Taking- out means for performing ,
Generating a plurality of kanji kana-mixed character string reading candidates by combining the reading extracted by the extracting means and kana candidate characters included in the candidate word , and converting the generated reading candidates to kana-kanji conversion Kana-kanji conversion means for performing processing
Collation means for performing a process of collating the kanji of the conversion result obtained by the kana-kanji conversion means and the candidate characters of the kanji part included in the candidate word;
As the candidate words to be processed in order from the first candidate word among the candidate words determined to be plural, the extraction unit, the kana-kanji conversion unit, and the collating unit perform respective processes, When the collation by the collating unit is successful, a candidate character as a recognition result is identified from among a plurality of candidate characters obtained by the recognizing unit, and the identified candidate character is recognized as a recognition result of the character string mixed with the kanji kana First output means for outputting as:
When it is determined that the number of candidate words formed by the candidate word forming unit is one, the candidate characters constituting the candidate word are identified, and the identified candidate characters are recognized as the kanji kana mixed character string. Second output means for outputting as a result.
[0009]
Here, according to a preferred embodiment of the present invention, the input means preferably inputs characters as stroke information, and in particular, the stroke information is preferably input by handwritten characters using a predetermined pen. As a result, if the stroke input surface is limited in size, has a large number of strokes, or forgets the kanji itself, it will be recognized by entering the kana with a high recognition rate. It becomes possible to make it.
[0012]
【Example】
Hereinafter, an implementation example of the present invention will be described in detail with reference to the accompanying drawings.
[0013]
FIG. 2 is a block diagram showing the configuration of an information processing system in which the character processing method of the present invention is implemented.
[0014]
The character string input from the input device 1 (handwriting pad, scanner, etc.) is processed by the central processing unit 2 (CPU), and the dictionary and its processing result are stored in the main storage device 4 (RAM, etc.) and the auxiliary storage device 5 ( Stored in a floppy disk, hard disk, etc.) and output by an output device 3 (CRT, printer, etc.).
[0015]
FIG. 1 is a flowchart showing the processing flow of the present invention. The program based on the flowchart is stored and held in the auxiliary storage device 5, and is loaded into the main storage device 2 and executed when the power is turned on.
[0016]
In this embodiment, the following steps are provided. That is,
Step S11 for inputting a reading character string from the input device 1, Step S12 for Kana-Kanji conversion, Step S14 for acquiring candidate readings, Step S15 for converting the acquired readings to Kana-Kanji, and the results and readings of Kana-Kanji conversion In step S16 for determining whether or not the extracted characters match, and in step S17 for acquiring the next data if they do not match, the determined characters are output to the output device 3 or the main storage device 4 and the auxiliary Step S18 is output to the storage device 5.
[0017]
Next, the flow of processing will be described based on FIG. 1, FIG. 6, and FIG. Here, a description will be given about the case where the type "gene" as a specific example.
[0018]
As shown in FIG. 6, the character string “Denko” is input (step S11). A known recognition process is performed from a dictionary stored in the main storage device 4 or the auxiliary storage device 5 (step S12).
[0019]
In the character recognition in the embodiment, the distance between the character vector of the character image or character stroke to be recognized and each standard vector stored in the recognition dictionary is calculated, and the character having a small calculated value is used as the character of the recognition result. Output. At this time, the first candidate character, the second candidate character,... Are in order of increasing distance calculation value, but the distance calculation value of the first candidate character is less than or equal to a predetermined value, that is, the probability is high. If there is a sufficient difference, only one candidate character is output. If this is not the case, a predetermined number of candidate character groups equal to or less than the second candidate (however, the calculation distance value is equal to or less than a predetermined threshold value) is output.
[0020]
Now, a character string candidate is created with the recognized characters obtained by the character recognition process. Here, if a plurality of candidate characters are output for one character in the recognition process, candidate character strings are created for the number of combinations. Therefore, in step S13, it is determined whether there are one or more candidate character strings, and it is determined whether the input character string has been identified.
[0021]
In the case of FIG. 6, since there are four candidate character strings, it cannot be specified. In this case, the process proceeds to step S14, the reading dictionary stored in the main storage device 4 or the auxiliary storage device 5 is searched, and the reading of the character string candidate is acquired . Acquire “Sen”, “Chi”, and “ぢ” for “1000”.
[0022]
Next, as shown in FIG. 7, kana-kanji conversion is performed on the reading acquired in step 14. For “Densen”, “Denchi”, and “Den ぢ” readings, conversion results such as “Electric wire”, “Infectious”, “Battery”, “Taji”, “Dentsu”, and “Tatsumi” are acquired. At this time, the conversion result is left so that the number of clauses of the conversion result is minimized or the reading is the longest.
[0023]
In step S16, it is determined whether or not the character “1000” is included in “wire”, “contamination”, “battery”, and “country” as the result of step S17. In this case, since this is not the case, the process proceeds to step S17, paying attention to the next candidate character string, for example, “denko”, and returns to step S14.
[0024]
Now, search the reading dictionary for “child” that performs the same process for “child” of “denko”, and search for “man”, “ko”, “go”, “shi”, “ji”. ”,“ Su ”,“ ne ”,“ mi ”readings. Candidates for reading handwritten text from acquired readings “Denotoko”, “Denko”, “Dengo”, “Denshi”, “Denji”, “Densu”, “Denne”, "Denmi" reading "den""male""denitsu""den""five""electronic""denshi""electromagnetic""tachi""den" “Nest”, “Transmission”, “Transmission”, “Body”, etc. are acquired as conversion results. As a result, “electricity”, “electron”, “spring”, and “electromagnetic” remain. Since there is an “electronic” that matches the “child”, the process proceeds from step S16 to step S18 . “Child” is determined as a recognized character, and “Denko” is output to the output device 3, the main storage device 4, or the auxiliary storage device 5 (step 18).
[0025]
As described above, according to the present embodiment, when a kanji kana mixed character string to be recognized is input, if there is a kanji in a combination of candidates obtained by recognition, the kanji is converted to “reading”. , Convert all to hiragana and kana-kanji conversion. Then, by collating this kana-kanji conversion and the result obtained by previously recognizing characters, it becomes possible to recognize recognition candidate characters with a high recognition rate and without increasing the size of the dictionary.
[0026]
[Other embodiments]
Next, a second embodiment will be described. In the second embodiment, as in the above embodiment, when the character string “Denko” is input, “Den Sen”, “Denko”, “ When "Nu" and "Denrai" are generated, "Electron" is output as output.
[0027]
Hereinafter, the flow of processing will be described with reference to FIGS.
[0028]
As shown in FIG. 6, the character string “Denko” is input (step 11). Recognized character candidates “densen”, “denko”, “denboshi”, and “denyu” are created from the dictionary stored in the main storage device 4 or the auxiliary storage device 5 (step 12). . Since there are four candidates for determining whether it can be specified, in this case, it cannot be specified (step 13). A reading dictionary stored in the main storage device 4 or the auxiliary storage device 5 is searched to obtain a reading of a character candidate. For “1000”, “Sen”, “Chi”, and “ぢ” are acquired (step 14).
[0029]
As shown in FIG. 8, kana-kanji conversion is performed on the reading acquired in step 14. For the readings of “Densen”, “Denchi”, “Den ぢ”, obtain conversion results such as “Electric wire”, “Infection”, “Battery”, “Taji”, “Transmission”, “ぢ” ( Step 15). As the number of clauses this time conversion result is minimized, or leaving the conversion result as readings ing the longest. As a result, “wire”, “contamination”, “battery”, and “tachi” remain, and “1000” is not included in the conversion result (step 16). The next data “child” is acquired (step 17).
[0030]
The same processing is performed for “child”. For “child”, the reading dictionary is searched, and “man”, “ko”, “go”, “shi”, “ji”, “su”, “ne”, “ne”, “ Get a reading of Candidates for reading handwritten text from acquired readings “Denotoko”, “Denko”, “Dengo”, “Denshi”, “Denji”, “Densu”, “Denne”, "Denmi" reading "den""male""denitsu""den""five""electronic""denshi""electromagnetic""tachi""den" “Nest”, “Transmission”, “Transmission”, “Body”, etc. are acquired as conversion results. As a result, it is determined that “electronic”, “electron”, “outgoing”, and “electromagnetic” remain, and there is “electron” that matches “child” (step 16 ). The character “electronic” determined as the optimum notation form as a result of the kana-kanji conversion is output to the output device 3, the main storage device 4, or the auxiliary storage device 5 with “child” as a recognized character (step 18).
[0031]
As described above, according to the second embodiment, it is originally desirable to be written in a kanji character string, but when a part of the kanji is input in hiragana, the word of the intended kanji character string is output. Will be.
[0032]
In the embodiment, the input character string can be applied to both cases where the character is captured and recognized as an image (such as OCR) and input using a handwriting pad (for example, a dedicated pen). For example, an image may be read and recognized from a storage medium storing the image, or an image (for example, a facsimile) received via a line may be directly recognized. In any case, as can be easily inferred from the above description, the effect is great when the present invention is applied to a device that directly enters characters by hand and recognizes them. This is particularly effective when the input area for pen input is somewhat narrow and it is difficult to input characters with a large number of strokes, or when the kanji itself cannot be remembered.
[0033]
In addition, when the input is handwritten input such as a handwriting pad, the recognition dictionary also becomes a dictionary for handwriting recognition, and although the detailed processing for recognizing characters is different, it is returned to the subsequent reading, kana-kanji conversion, and the recognized characters Since the process for determining the column can be performed in the same manner, the effect is great.
[0034]
Furthermore, although not specifically described in the embodiment, when the kana-kanji conversion is performed on the reading of the candidate word and the matching process is performed, the characters constituting the candidate word are the first candidate characters obtained by the recognition process. It is desirable to use them in order from the candidate. The reason is that the number of loops in steps S14 to S17 in FIG.
[0035]
Furthermore, as can be easily inferred from the above description, the present invention may be applied to a system composed of a plurality of devices or an apparatus composed of a single device. Needless to say, the present invention can also be applied to a case where the present invention is achieved by supplying a program to a system or apparatus.
[0036]
【The invention's effect】
As described above, the data processing method of the present invention includes a step of inputting a character string, a step of recognizing a character, a step of extracting a kanji reading from the dictionary, a step of converting the kana-kanji character, and a converted character A step of drawing a recognized character candidate and a step of outputting the recognized character string, thereby improving the recognition rate of character recognition and sharing the storage capacity with the Kana-Kanji conversion dictionary. There are effects such as reduction.
[0037]
[Brief description of the drawings]
FIG. 1 is a flowchart showing a processing procedure of character recognition processing in an embodiment.
FIG. 2 is a block diagram of a system according to the embodiment.
FIG. 3 is a flowchart showing a conventional character recognition processing procedure.
FIG. 4 is a diagram for explaining an outline of conventional character recognition;
FIG. 5 is a diagram for explaining an outline of conventional character recognition;
FIG. 6 is a diagram showing an outline of character recognition processing in the embodiment.
FIG. 7 is a diagram showing an outline of character recognition processing in the embodiment.
FIG. 8 is a diagram showing an outline of character recognition processing in the second embodiment.
[Explanation of symbols]
1 Central processing unit 2 Main storage device 3 Input device 4 Display device 5 Auxiliary storage device

Claims (6)

An input means for inputting a kanji mixed character string;
Character recognition means for recognizing characters of a kanji mixed character string input by the input means, and outputting one or more candidate characters for each character;
Candidate word forming means for forming one or a plurality of candidate words for the character string mixed with kanji by combining candidate characters of each character obtained by character recognition by the character recognition means;
The candidate If word candidate word formed by formation means determines that there are multiple, the readings of the candidate Chinese characters included in the candidate word that has been processed from among the plurality of candidate words to one or more retrieving process Taking- out means for performing ,
Generating a plurality of kanji kana-mixed character string reading candidates by combining the reading extracted by the extracting means and kana candidate characters included in the candidate word , and converting the generated reading candidates to kana-kanji conversion Kana-kanji conversion means for performing processing
Collation means for performing a process of collating the kanji of the conversion result obtained by the kana-kanji conversion means and the candidate characters of the kanji part included in the candidate word;
As the candidate words to be processed in order from the first candidate word among the candidate words determined to be plural, the extraction unit, the kana-kanji conversion unit, and the collating unit perform respective processes, When the collation by the collating unit is successful, a candidate character as a recognition result is identified from among a plurality of candidate characters obtained by the recognizing unit, and the identified candidate character is recognized as a recognition result of the character string mixed with the kanji kana First output means for outputting as:
When it is determined that the number of candidate words formed by the candidate word forming unit is one, the candidate characters constituting the candidate word are identified, and the identified candidate characters are recognized as the kanji kana mixed character string. A character recognition device comprising: a second output means for outputting as a result.   The character recognition apparatus according to claim 1, wherein the character string input by the input unit is a character string input by handwriting.   The character recognition apparatus according to claim 1, wherein the character string input by the input unit is a character image. A character recognition step of recognizing characters of a character string mixed with kanji input and outputting one or more candidate characters for each character;
A candidate word forming step of forming one or more candidate words for the character string mixed with kanji by combining candidate characters of each character obtained by character recognition in the character recognition step;
If the candidate word formed by said candidate word forming step have multiple fetches for performing reading one or more retrieving process of candidate Chinese characters included in the candidate word that has been processed from among the plurality of candidate words Process,
Generating a plurality of kanji kana-mixed character string reading candidates by combining the reading extracted in the extraction step and kana candidate characters included in the candidate word , and converting the generated plurality of reading candidates to kana-kanji conversion A kana-kanji conversion process for performing
A collation step of performing a process of collating the kanji of the conversion result obtained in the kana-kanji conversion step and the candidate characters of the kanji part included in the candidate word;
As the candidate words to be processed in order from the first candidate word among the candidate words determined to be plural, the respective processes by the extraction step, the kana-kanji conversion step, and the matching step are performed, When the collation in the collation process is successful, a candidate character as a recognition result is identified from among a plurality of candidate characters obtained in the recognition step, and the identified candidate character is recognized as a character string mixed with the kanji kana A first output step to output as:
When there is one candidate word formed in the candidate word formation step, a candidate character constituting the candidate word is specified, and the specified candidate character is output as a recognition result of the character string mixed with the kanji kana. A character recognition method comprising: an output step.   The character recognition method according to claim 4, wherein the input character string mixed with kanji characters is a character string input by handwriting.   5. The character recognition method according to claim 4, wherein the inputted character string mixed with kanji characters is a character image.

Images