JP2000075887A - Device, method and system for pattern recognition - Google Patents

Abstract

PROBLEM TO BE SOLVED: To increase the recognition object patterns and to reduce the network load. SOLUTION: In the pattern recognition system, a terminal 10, which is provided with a small scale dictionary 15 that becomes a reference for pattern recognition, and a server 20, which is provided with a large scale dictionary 24 having a larger scale than the dictionary 15, are connected by a network. The terminal 10 conducts a recognition process for the pattern inputted into an input section 11 using the dictionary 15. If the recognition result is not appropriate, recognition is conducted by the dictionary 24 of the server 20.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method using pattern recognition, and more particularly to an apparatus, method and system for pattern recognition in a plurality of devices connected by a network.

[0002]

2. Description of the Related Art Generally, when performing pattern recognition,
The larger the pattern dictionary in which the reference patterns are registered for matching, the more detailed matching can be performed, the higher the recognition accuracy, and the more types of patterns can be recognized. Becomes

[0003] On the other hand, when the pattern dictionary is made large, the amount of calculation and storage capacity required for the recognition process become large.

[0004] Therefore, even in a normal apparatus, the response of the recognition processing cannot be deteriorated.

[0005] In addition, especially when the calculation amount and the storage capacity are severely restricted, such as in a small terminal that is assumed to be portable, it has been difficult to realize pattern recognition using a large-scale pattern dictionary by itself.

A method for performing pattern recognition processing using a large-scale pattern dictionary on such a small terminal is as follows.
As shown in the specification disclosed in JP-A-7-152797, a server is connected to a server via a network, and the server performs a pattern recognition process using a large-scale pattern dictionary. Is sent to a terminal via a network.

[0007] That is, the above specification discloses one configuration example of the information access system as shown in FIG. This information access system comprises a center system 101 and, for example, a PSTN or an integrated services digital network (IS).
The terminal 102 is connected by a telephone line such as DN) and includes a terminal 102 that can communicate with the center system 101.

The terminal 102 is connected to the center system 1
01 voice input type database search / transmission system 10
3 is used as an interface to access a book database 105 in which information of publications such as library information, newspaper information, magazine information, and new book information is stored, and to obtain desired information. The book database 105 stores, as information on publications, phonetic symbol information including phonemic information and prosodic information in addition to text data and the like.
In No. 2, in addition to receiving such information, displaying and printing the text data of the publication, and generating a synthesized sound from the phonetic symbol information, it is possible to obtain a speech reading the publication. .

The center system 101 comprises a voice input type database search / transmission system 103, an authoring system 104, a book database 105, and a network control unit 106. The authoring system 104 is configured to store publication information in the book database 5, create a database of publication information, and edit the information stored in the book database 105. Network control unit 10
6 controls communication between the center system 101 and other center systems, and
Of the book database 5 and databases in other center systems.

[0010]

However, pattern recognition processing using a large-scale pattern dictionary has a problem of processing response.

That is, when the size of the pattern dictionary for recognition is increased, it takes time to collate the input pattern, it takes a long time until a recognition result is obtained, and the response becomes poor. Therefore, the operability of the device may be degraded.

For example, consider a terminal that uses voice recognition processing, which is an example of pattern recognition processing. Poor response in recognizing commonly used words, such as commands for controlling the terminal, greatly impairs the operability of the terminal.

Further, in the case of a pattern recognition processing apparatus configured to realize a pattern recognition process using a large-scale pattern dictionary using a terminal and a server as in the information access system shown in FIG. There is such a problem.

That is, to execute pattern recognition in a terminal, the terminal first connects to a server via a network, transfers data of an input pattern, performs pattern recognition processing in the server, and returns the result via the network again. It is necessary to perform the process of obtaining from the server. Therefore, it takes time to obtain a pattern recognition result, and the response is further reduced.

Further, in the case where the pattern recognition is always performed in the server, a terminal using the network must be connected to the server every time the pattern recognition is performed, so that the load on the network increases. In an environment that is not connected, the cost of using the network increases.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and has as its object to provide a pattern recognition apparatus and method which can shorten the response of pattern recognition processing and increase the number of patterns to be recognized. .

[0017]

In order to solve the above-mentioned problems, a pattern recognition apparatus according to the present invention comprises: an input means for inputting a pattern; a first dictionary used as a reference for pattern recognition; A second dictionary, which is a reference dictionary used for recognition of the first dictionary and which is larger in scale than the first dictionary, and a pattern input to the input means is recognized by the first dictionary. In this case, there is provided a recognizing means for performing recognition using the second dictionary, and an output means for outputting a result of pattern recognition from the recognizing means.

Further, the pattern recognition method according to the present invention comprises:
An input step of inputting a pattern, a first dictionary serving as a reference used for pattern recognition, and a second dictionary serving as a reference used for pattern recognition and being larger than the first dictionary, A recognition step of recognizing the pattern input to the input step by the first dictionary and recognizing that the result is inappropriate by the second dictionary; and a pattern recognition result from the recognition step. And an output step of outputting

The pattern recognition system according to the present invention includes a terminal having a first dictionary serving as a reference used for pattern recognition, and a terminal serving as a reference serving as a reference used for pattern recognition and having a scale larger than that of the first dictionary. Is a pattern recognition system in which a server having a second dictionary having a large size is connected via a network, wherein the terminal recognizes a pattern input to the terminal using the first dictionary,
If the result is inappropriate, recognition is performed using the second dictionary of the server.

That is, in the present invention, a pattern dictionary serving as a reference for a matching process in pattern recognition is stored in a small first
And a large-scale second dictionary, which are stored separately, and used as needed.

That is, pattern recognition processing is first performed using a small first dictionary. If an appropriate recognition result is obtained, the pattern recognition processing ends. If an appropriate result is not obtained, the pattern recognition processing is performed again using the large-scale second dictionary.

[0022]

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

First, as a first embodiment of the present invention,
A speech recognition device that recognizes speech by pattern recognition will be described.

As shown in FIG. 1, the voice recognition device includes a terminal 10 on which data input / output operation is performed by a user;
And a server 20 having a large-capacity database and centrally performing data processing. These terminals 10
The server 20 and the server 20 can be connected by a network 30 which is a line for transmitting information.

First, the terminal 10 will be described. Terminal 1
0 is for accepting data input / output by the user and connecting to the server 20 via the network 30.

The terminal 10 includes an input unit 11 for receiving a voice input, a recognition dictionary unit 12 corresponding to a pattern dictionary,
The operation unit 1 for collating the voice pattern by the recognition dictionary unit 12
3 and an output unit 1 for outputting a result sent from the arithmetic unit 13
And 4.

The input section 11 is a section for receiving an input of a voice pattern to be recognized, and includes a microphone, an amplifier, an AD converter, and the like. The input unit 11 digitizes the input voice pattern and sends it to the arithmetic unit 13.

The recognition dictionary unit 12 corresponds to a pattern dictionary. The recognition dictionary unit 12 includes a semiconductor memory, a hard disk, and the like. In the case of speech recognition, the recognition dictionary unit 12 generally includes an acoustic pattern that is a basic unit of recognition, a word dictionary that defines how acoustic patterns can be linked, and a grammar.

In the present embodiment, the recognition dictionary unit 12
Are implemented in a small dictionary 15 which is a small recognition dictionary. The small dictionary 15 includes an acoustic pattern 16 and a word dictionary / grammar 17.

The present embodiment also includes a large-scale dictionary, which is a recognition dictionary having a larger scale than a small-scale dictionary. This large-scale dictionary has a sound pattern, a large vocabulary word dictionary, and a complicated grammar that are more detailed than the small-scale dictionary 15. As will be described later, this large-scale dictionary is provided in the server 20, so that the terminal 10 can be connected to the network 30 via the network 30.

The operation unit 13 is composed of a CPU, a semiconductor memory and the like. The arithmetic unit 13 collates the voice pattern sent from the input unit 11 with the acoustic pattern stored in the recognition dictionary unit 12 in a form that satisfies the connection conditions of the pattern specified in the word dictionary and the grammar. The calculation unit 13 sends the speech recognition result obtained by the speech recognition based on the collation to the output unit 14.

The output unit 14 is composed of a display, a voice synthesizer, and the like. The output unit presents the user with the speech recognition result sent from the calculation unit 13.

The terminal 10 has a server connection unit 18 which is an interface unit for connecting to the network 30. The server connection unit 18 is configured by a modem or the like,
Connect to the server via the network as needed.

Next, the server 20 will be described. The server 20 has a large-capacity database, and performs data processing in a concentrated manner by being connected from a terminal via the network 30.

The server 20 has a recognition dictionary unit 21 corresponding to a pattern dictionary, an operation unit 22 for recognizing a voice pattern using the recognition dictionary unit 21, and a terminal connection unit 23 which is an interface for connecting to a network 30. are doing.

The recognition dictionary unit 21 has substantially the same configuration as the recognition dictionary unit 12 provided in the terminal 10. That is, the recognition dictionary unit 21 corresponds to a pattern dictionary,
It is composed of a semiconductor memory and a disk. However, a large-scale dictionary 24 which is larger than the small-scale dictionary 15 provided in the terminal 10 is provided. In the large-scale dictionary 24,
An acoustic pattern 25, a word dictionary of a large vocabulary, and a complicated grammar 26 are provided.

The operation unit 22 has substantially the same configuration as the operation unit 13 provided in the terminal 10. That is, the operation unit 22
Is a connection between a voice pattern sent from the terminal 10 via the network and an acoustic pattern stored in the recognition dictionary unit 21 by a CPU, a semiconductor memory, and the like. Perform matching in a form that satisfies the conditions. However, the operation unit 2 provided in the server 20
For 2, a unit having a higher performance than the operation unit 13 provided in the terminal 10 is used. The operation unit 22 includes a recognition dictionary unit 21
The speech recognition result obtained by the speech recognition based on the collation using is transmitted to the terminal 10.

The terminal connection unit 23 is composed of a modem or the like. The terminal connection unit 23 receives a server connection command from the terminal 10 and controls connection with the terminal 10 based on the server connection command.

The frequently used patterns are registered in the small dictionary 15, which is a small recognition dictionary, so that the frequently used words can be quickly recognized by the small dictionary 15. It can be performed.

Further, a large-scale recognition dictionary 2
By providing four dictionaries on the server 20 and using them via the network 30, it is also possible to perform recognition with more patterns and more detailed matching. In addition, since this is executed only when necessary, the load on the network 30 and the communication cost can be reduced as compared with the case where the connection processing is always performed by the network 30 and the server 20 performs the recognition processing.

Next, the flow of the voice recognition process in the voice recognition device will be described with reference to the flowchart shown in FIG.

In the first step S11, the terminal 1
A voice is input to the input unit 11 of “0”. Input unit 1
In step 1, the input voice is digitized and sent to the operation unit 13.

In step S12, the operation unit 13
Performs a recognition process on the digitized voice sent from the input unit using the small dictionary 15 of the recognition dictionary unit 12 of the terminal 10. That is, recognition processing is performed using the acoustic pattern 16 and the word dictionary / grammar 17 provided in the small-scale dictionary 12.

In step S13, step S1
The result of the speech recognition process performed in step 2 is determined. There are two methods for determining the result of the recognition process, a method automatically performed by the voice recognition device and a method of confirmation by the user.

As a method of automatically performing the result of speech recognition by the speech recognition apparatus, generally, a confidence measure (confidence measu) is used.
There is a method of making a judgment based on a measure relating to the reliability of the recognition result called re; C).

Here, the reliability scale C includes the recognition score P (Wfirst) of the candidate having the first recognition score and the recognition score P (Wsecond) of the candidate having the second recognition score.
The value calculated as follows is used. C = P (Wfirst) −P (Wsecond) Further, the reliability scale C is obtained by using the recognition score P (Wthird) of the third candidate or less, and C = W (Pfirst) / (P (Wsecond) − P (Wthir
d)) can also be given by

In general, the reliability scale of the above-described reliability measure is such that the larger the difference between the recognition scores of the first and second or lower candidates as a result of the pattern recognition processing, the higher the reliability of the recognition result. It is defined based on what is assumed.

According to such a method of making a determination based on the reliability measure, the speech recognition apparatus calculates the reliability measure in the course of the speech recognition processing, and determines whether the value of the reliability measure exceeds a predetermined threshold value. Determines that the recognition result is OK, and if the value of the reliability scale is smaller than the threshold value, the recognition result is N
It is determined to be G.

On the other hand, in the method based on confirmation by the user,
The result of the recognition processing is sent from the calculation unit 13 to the output unit 14. Then, in the output unit 14, the result is transmitted to the user by screen display or audio presentation, and the result is determined by the user, and the quality of the recognition result is determined.

In step S14, step S1
The process branches depending on the judgment of the recognition result in 3. That is,
If the recognition result is determined to be OK, step S20
Then, the recognition result is sent to the output unit 14 as a final result output as a final result, and is output to the user, and the process ends. If it is determined that the recognition result is NG, the process proceeds to the next step S15.

In step S15, the terminal 10 connects to the server from the server connection unit 18 via the network 20. Then, the terminal 10 transfers data necessary for voice recognition to the server 20. Here, the data necessary for speech recognition may be the input speech data itself,
The results of the acoustic analysis and the results of the vector quantization can also be transferred. By transmitting the result of voice analysis and the result of vector quantization, the amount of information to be transferred can be reduced. In the server 20, the transferred data is received by the terminal connection unit 23, and the data is
Send to

In step 16, the arithmetic unit in the server 20 transmits the data sent from the terminal 10 to the recognition dictionary unit 21.
Is compared with the data held in the large-scale dictionary 24 of
Perform recognition. Then, the arithmetic unit 22 sends the result of the speech recognition using the large-scale dictionary 24 from the terminal connection unit 23 to the terminal 10 via the network 30. Last step S20
In, the terminal 10 outputs the recognition result sent from the server as a final result.

At the time of recognition by the server 20 using the large-scale dictionary 24, the following processing can be performed at the same time.

That is, in S17, the large dictionary 2
Regarding each pattern in No. 4, especially word recognition, when obtaining a recognition result by referring to the large-scale dictionary 24, how many times the pattern has been used as a recognition result in the past is stored. Each time recognition is performed with reference to the large-scale dictionary, 1 is added to the number of times the pattern obtained as a result of recognition is used.

If, in step S18, the number of past uses exceeds a certain threshold, or if the frequency of uses in a certain period exceeds a certain threshold, the corresponding pattern is sent from the server to the terminal in step S19. Transfer and register in the small dictionary in the terminal.

As described above, by transferring frequently accessed patterns to the small dictionary 15 of the terminal 10, the number of times the terminal 10 accesses the server 20 can be reduced. Further, it is effective in improving the response of the recognition process and reducing the load on the network 30.

The registration of the pattern in the small-scale dictionary 15 of the terminal 10 can be automatically performed by the voice recognition device as described above. It is also possible.

Although not registered in the small dictionary 15,
Of the words registered in the large-scale dictionary 24, words frequently used by the user are also registered in the small-scale dictionary 15, so that the number of accesses to the large-scale dictionary 24 can be reduced. 30
The load can be improved.

Next, an example of actual processing of the speech recognition apparatus will be described. As an example of actual processing of the voice recognition device, it is conceivable to incorporate the above-described voice recognition device into a database search by the terminal 10. For example, consider a terminal 10 that searches for book information and a server 20 that waits for book information. Commands for controlling the terminal 10, such as clearing and scrolling of drawings, are stored in the small dictionary 15, and those related to the book information such as book names, author names, and publisher names, and those having a large number are stored in the large-scale dictionary 24. Register with.

The user uses the small dictionary 15 built in the terminal 10 for frequently used commands such as clearing the screen.
, The terminal can be operated with a quick response without using the network 30.
If the user inputs a name that is not registered in the small dictionary 15 such as a book name, the terminal 10 automatically connects to the server 20 via the network 30 and executes a recognition process. The terminal can be used without being aware that the dictionaries are distributed and connected to the network.

At this time, by determining the communication protocol between the terminal 10 and the server 20, the names of the control commands of the individual devices can be determined independently for each device.

Next, as a second embodiment of the present invention,
A handwritten character recognition device that recognizes handwritten characters by pattern recognition will be described.

First, the configuration of the handwritten character recognition device will be described with reference to FIG. In FIG. 3, parts corresponding to those of the speech recognition apparatus shown in FIG. 1 are denoted by the same reference numerals.

The speech recognition device is composed of a terminal 10 on which data input / output operation is performed by a user, and a server 20 having a large-capacity database and performing centralized data processing. These terminal 10 and server 20
Can be connected by a network 30 which is a line for transmitting information.

The terminal 10 receives data input and output by the user and connects to the server 20 via the network 30. The terminal 10 includes an input unit 11 for receiving a voice input and a recognition dictionary unit 1 corresponding to a pattern dictionary.
2, an operation unit 13 for collating the voice pattern by the recognition dictionary unit 12, an output unit 14 for outputting the result sent from the operation unit 13, and a server connection unit 18 for connecting to the server 20 via the network 30. Have.

The input unit 11 for receiving an input of a handwritten character pattern to be recognized is constituted by a touch pad or the like. The input unit 11 digitizes the input handwritten character pattern and sends it to the calculation unit 13.

The recognition dictionary unit 12 provided with the small-scale dictionary 15 corresponding to the pattern dictionary stores a character pattern 19 as a basic unit of recognition and a word dictionary / grammar 17 defining how the character pattern 19 is formed. Have.

Note that the configuration of the other parts of the terminal 10 and the processing of recognition in the terminal 10 are the same as those of the above-described speech recognition apparatus, and a description thereof will be omitted.

The server 20 has a recognition dictionary unit 21 corresponding to a pattern dictionary, an operation unit 22 for recognizing a voice pattern using the recognition dictionary unit 21, and a terminal connection unit 23 which is an interface for connecting to the network 30. are doing.

The large dictionary 24 provided in the recognition dictionary unit 21 and having a larger size than the small dictionary 14 is
5 has a finer character pattern 19, a large vocabulary word dictionary, and a complicated grammar 26.

The configuration of the other parts of the server 20 and the processing of recognition in the server 20 are the same as those of the above-described speech recognition apparatus, and therefore the description is omitted.

As described above, in the handwritten character recognition device of the present embodiment, the small-scale dictionary 15 and the large-scale dictionary 2
4 that the character pattern is recognized in place of the acoustic pattern, and the input unit 11 is configured with a touch pad,
The digitization of the input handwritten character is different from the voice recognition device of the first embodiment described above.

The processing procedure in the handwritten character recognition apparatus is basically the same as the processing procedure in the voice recognition apparatus shown in FIG. 2, except that a character pattern is processed instead of a voice pattern.

That is, the processing steps of the handwritten character recognition device will be described with reference to FIG. 2. In the first step S11, the terminal 10 replaces the voice input with the input unit 1
1. Input handwritten characters on the touch pad or the like. In the next step S12, the "operation unit 13
The small dictionary 1 of the recognition dictionary unit 12 recognizes the input handwritten characters.
In step 5, the character pattern 19 and the word dictionary / grammar 17 are used for recognition.

In step S13, step S1
The result of the recognition in 2 is determined. As described above, the determination of the recognition result is performed, for example, using a confidence measure (confidence m).
There is a method of automatically performing the determination based on the easure) and a method of making a determination by presenting to the user. If the determination of the recognition result is OK, the process proceeds to step S20, and the final result is output to the output unit 14.
And exit. If the determination of the recognition result is NG, the process proceeds to step S15.

In step S15, the terminal 10
The server 30 connects to the server 30 via the network 30 via the server connection unit 18 and sends the input handwritten characters to the server 20. Here, the handwritten character data sent to the server 20 does not need to be the input handwritten character itself, and the amount of information required for transfer can be reduced by sending the result of analysis or the result of appropriate quantization. What can be recommended is the same as the above-described speech recognition device.

In step S 16, in the server 20, the calculation unit 22 determines the character pattern 27 and the word dictionary / grammar 26 of the large-scale dictionary 24 included in the recognition processing unit 21.
Is used to recognize handwritten characters. And the server 20
Sends the result of the recognition process to the terminal 10 via the network 30 by the terminal connection unit 23. Then, step S2
In the case of 0, the terminal displays the result recognized by the server on the output unit 14 and ends the series of steps.

The number of times the recognition result is used is measured, and words registered in the large-scale dictionary 24 are registered in the small-scale dictionary according to the frequency of access, whereby the server 20 of the terminal 10
It is also possible to propose the number of accesses to the same as in the above-described speech recognition device. Such a large-scale dictionary 2
The registration of the word from No. 4 to the small dictionary 15 is shown in steps S17 to S19, but is the same as in the first embodiment described above, and will not be described.

As described above, the processing procedure in the handwritten character recognition device is either the processing procedure in the voice recognition device and the processing of the pattern recognition algorithm for voice recognition performed by acoustic pattern recognition, or the handwriting character recognition. It is different for handwritten character recognition.

Next, a description will be given of a process performed by the handwritten character recognition apparatus.

In the handwritten character recognition device, information is obtained by voice from the user in the voice recognition device, but the information is changed to input by handwritten characters. This makes it difficult for ordinary terminals 10 to use difficult kanji in personal names and place names.
Characters that are not registered in the small dictionary 15 of
By connecting to the large-scale dictionary 24 via the network 30, recognition becomes possible.

In the configuration of the handwritten character recognition device shown in FIG. 3, a configuration in which the small-scale dictionary 15 and the large-scale dictionary 24 are distributed and arranged in the terminal 10 and the server 20 is shown.
However, both the small dictionary 15 and the large dictionary 24 can be built in the same device. In that case, by performing recognition using the small-scale dictionary 15 first, those frequently input can be recognized with a quick response.

Also, in the speech recognition apparatus of the first embodiment, by providing the small dictionary 15 and the large dictionary 24 in the same apparatus, those which are frequently input are recognized with a quick response. be able to.

As described above, according to the present embodiment, the recognition dictionary used for pattern recognition is divided into small ones and large ones, and the dictionary is used for recognition processing by appropriately using them. To shorten the response and increase the number of patterns to be recognized. Furthermore, by providing a small dictionary in a terminal and a large dictionary in a server connectable to a network, it is possible to increase the number of recognition target patterns and reduce the network load in a small terminal.

[0085]

According to the present invention, by using a small-sized pattern dictionary and a large-sized pattern dictionary together, frequently used patterns are registered in the small-sized dictionary, and the response of the pattern recognition is accelerated. It is possible to reduce the number of accesses to a large-scale dictionary. Thus, when pattern recognition using a large-scale dictionary is provided by a server that can be connected via a network, compared to the case where pattern recognition is always performed using a server,
Network load can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a schematic configuration of a speech recognition device.

FIG. 2 is a flowchart illustrating a processing procedure in the voice recognition device.

FIG. 3 is a block diagram illustrating a schematic configuration of a handwritten character recognition device.

FIG. 4 is a block diagram illustrating a configuration of an example of an information access system.

[Explanation of symbols]

10 terminals, 15 small dictionaries, 20 servers, 24
Large dictionary, 30 networks

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Honda, etc. 6-7-35 Kita Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Osamu Hamada Osamu Hamada 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo Sony Corporation F term (reference) 5B064 DA15 DA27 DA29 5D015 HH13 LL10 9A001 HH17 HH21 HH22

Claims (16)

[Claims] An input unit for inputting a pattern; a first dictionary serving as a reference used for pattern recognition; and a second dictionary serving as a reference used for pattern recognition, the second dictionary being larger in size than the first dictionary. Dictionary, recognition processing means for recognizing patterns, determination processing means for determining whether or not the result of pattern recognition is appropriate, output means for outputting pattern recognition results, and input to the input means The recognized pattern is recognized by the recognition processing means using the first dictionary, and the result of recognition by the recognition processing means is determined by the determination processing means;
If the determination processing means determines that the recognition result is appropriate, the output means outputs the recognition result. If the determination processing means does not determine that the recognition result is appropriate, the pattern is recognized. A pattern recognition device, comprising: control means for making a determination using the second dictionary in the processing means and controlling the recognition result to be output in the output means. 2. The method according to claim 1, wherein said determination processing means automatically determines whether or not the recognition result obtained by said recognition processing means is appropriate, using a scale obtained in a process of pattern recognition. Pattern recognition device. 3. The pattern recognition apparatus according to claim 2, wherein said recognition processing means employs a reliability measure as said measure. 4. The pattern recognition device according to claim 1, wherein a pattern used for recognition in said second dictionary is registered in said first dictionary. 5. The method according to claim 1, wherein the registration of the pattern recognized by the second dictionary in the first dictionary is automatically performed in accordance with the frequency of reference to the second dictionary. The pattern recognition device according to claim 4, wherein 6. The pattern recognition apparatus according to claim 4, wherein no pattern is registered in the first dictionary in an initial state. 7. The pattern recognition apparatus according to claim 1, wherein said recognition processing means recognizes at least one of a voice, an image, and a character. 8. An input means for inputting a pattern, a first dictionary serving as a reference used for pattern recognition, and a second dictionary serving as a reference used for pattern recognition, the second dictionary being larger in size than the first dictionary. A recognition processing means for recognizing a pattern; an output means for outputting data; a confirmation means for receiving confirmation of a result of pattern recognition; and a first processing means for recognizing a pattern inputted to the input means in the first recognition processing means. And the recognition result in the recognition processing means is judged by the judgment processing means.
The result of the judgment by the judgment processing means is presented to the confirmation means. When the confirmation result is confirmed by the confirmation means, the recognition result is outputted by the output means, and the confirmation result is confirmed by the confirmation means. A recognition unit that uses the second dictionary to recognize that the recognition result has not been received, and a control unit that controls the output unit to output the recognition result. 9. The system according to claim 8, wherein said control means presents to said presenting means a scale obtained in the course of pattern recognition by said recognition processing means together with a result of the recognition processing by said recognition processing means. Pattern recognition device. 10. The pattern recognition apparatus according to claim 9, wherein said recognition processing means adopts a reliability scale as said scale. 11. The pattern recognition apparatus according to claim 8, wherein a pattern used for recognition in said second dictionary is registered in said first dictionary. 12. The registration of a pattern recognized by the second dictionary in the first dictionary is automatically performed according to the frequency of reference to the second dictionary. The pattern recognition device according to claim 11, wherein 13. The pattern recognition apparatus according to claim 11, wherein a pattern is not registered in the first dictionary in an initial state. 14. A first reference used for pattern recognition.
And a dictionary for pattern recognition,
An input step of inputting a pattern, a recognition processing step of recognizing the pattern, and a judgment of judging whether or not the result of the pattern recognition is appropriate, using a second dictionary that is larger than the first dictionary. A processing step, an output step of outputting a recognition result of the pattern, and a pattern input to the input step is recognized using the first dictionary in the recognition processing step, and the recognition result in the recognition processing step is determined. Judge in the processing process,
If the recognition result is determined to be appropriate in the determination processing step, the recognition result is output in the output step. If the recognition result is not determined to be appropriate in the determination processing step, the pattern is recognized. A control step of making a determination in the processing step using the second dictionary and controlling the recognition result to be output in the output step. 15. A first reference used for pattern recognition.
An input step of inputting a pattern using a dictionary which is a reference dictionary used for pattern recognition and a second dictionary which is larger than the first dictionary, a recognition processing step of recognizing the pattern, An output step of outputting data; a confirmation step of receiving confirmation of a pattern recognition result; a pattern input to the input step is recognized using the first dictionary in the recognition processing step; The recognition result is determined in the above determination processing step,
The judgment result in the judgment processing step is presented to the confirmation step. When the recognition result is confirmed in the confirmation step, the recognition result is output in the output step, and the recognition result is confirmed in the confirmation step. A recognition step using the second dictionary in the recognition processing step if the recognition is not received, and a control step of controlling the output step to output the recognition result. 16. An input means for inputting a pattern, a first dictionary used as a reference for pattern recognition, first recognition processing means for recognizing a pattern using the first dictionary, and pattern recognition. A terminal having judgment processing means for judging whether or not the result of the pattern is appropriate, output means for outputting a pattern recognition result, and control means for controlling pattern recognition; and a reference used for pattern recognition. A second dictionary, which is a dictionary larger than the first dictionary, and a server having second recognition processing means for performing a pattern recognition process using the second dictionary are connected by a network; A pattern recognition system, wherein the control means recognizes the pattern input to the input means using the first dictionary in the first recognition processing means, The recognition result in the recognition processing means is judged by the judgment processing means, and when the judgment processing means judges that the recognition result is appropriate, the recognition result is outputted by the output means, and the judgment result is outputted by the judgment processing means. If the recognition result is not determined to be appropriate, the pattern is determined by the second recognition processing means using the second dictionary, and control is performed so that the recognition result is output by the output means. And a pattern recognition system.