JP3267064B2 - Pattern information processing device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information processing apparatus for various patterns such as characters and voices, which is constructed on the premise of multi-user, and more particularly to an apparatus having a learning function.

[0002]

2. Description of the Related Art Conventionally, there has been proposed an apparatus which provides a system for learning on the assumption of a specific user, that is, a single user, such as an apparatus described in Japanese Patent Application Laid-Open No. 55-13973. That is, for each user, learning of the pattern given by each user ("learning"
This is realized by processing such as registration of predetermined data in a database or the like).

For this reason, in a group in which a plurality of users are members, even if a pattern is a common arrangement and commonly used by members in the group, the pattern commonly used is The pattern learning was performed individually for each user.

That is, when there is a mark or the like commonly used in a group, in order to be able to input by hand, each user who is a member of the group has to learn the pattern of the mark and the like for each individual. Had to be done individually.

[0005] Further, taking the kana-kanji conversion in a word processor as an example, as a kana-kanji conversion that is implicitly common within a group, for example, there is a kana-kanji conversion that converts "Bishamon" to "Bishamon". However, this fixed phrase registration had to be performed individually.

That is, in order to improve the operability of one's own pattern information processing system, it was necessary to perform pattern learning by oneself or by relying on a person. That is, the operability was not improved forever unless the pattern learning of the own pattern information processing system was performed by himself or by asking a person.

[0007]

Therefore, in order to improve operability without performing pattern learning of one's own pattern information processing system by oneself or by relying on a person, the same group must be used. The challenge is to use the result of a person who has learned the pattern of the person's pattern information processing system.

[0008] By learning implicitly (or explicitly specified) common rules within a group, individual operations of group members are omitted and operability is improved. That is, in the pattern information processing system for multi-users, learning of a pattern generally used not only by a specific user but also by a plurality of users is performed.

[0009] To this end, extracting a pattern that will be commonly used in a certain group composed of a plurality of users and learning the pattern as a pattern belonging to the group. Is an issue.

[0010] Conventionally, in a file system such as UNIX, it is possible to allow all group members to read / write by assigning attributes to files. However, with this file system technology alone, the system administrator in the group considers how to rewrite “what and how” to a file, that is, to improve operability by learning from a dictionary. There was a problem that the file had to be rewritten.

An object of the present invention is to provide an apparatus capable of learning implicitly (or explicitly) a common arrangement within a group without such a system administrator.

[0012]

The first feature of the present invention is as follows.
In a pattern information processing apparatus having at least an input device for receiving a given pattern and a dictionary having a configuration in which data is classified into a plurality of types and registered for a plurality of users, an attribute belonging to each user is set to A dictionary is registered, and a group dictionary and a dictionary editing device are provided. When a pattern given via the input device satisfies a predetermined condition, the dictionary editing device edits data in the dictionary. , The pattern and the code corresponding to the pattern are registered in the group dictionary, and users having the same belonging attribute can access the group dictionary.

More specifically, an input device for inputting a pattern, a dictionary in which the pattern and the category, affiliation, etc. to which the pattern belongs, and affiliation attributes (user 1, user 2, group 1, group 2) , Everyone, etc.)
Receiving an input pattern and an attribute belonging to an attribute designating device, searching for a pattern having a corresponding attribute belonging to the pattern registered in the dictionary, and searching for a pattern most similar to the input pattern, A recognition device that outputs the category, and a dictionary editing device that extracts a pattern commonly used in a certain group and performs editing processing such as registering the pattern as a group to which the user belongs as a group to which the user belongs in a common dictionary. It has the structure which has.

A second feature of the present invention is that a plurality of input devices for receiving at least a given pattern,
In a pattern information processing apparatus having, for each of the input devices, a dictionary for storing the input pattern and one or a plurality of codes corresponding to the input pattern together with priorities, a code corresponding to a pattern given via a certain input device , Searching from a dictionary corresponding to the input device and a plurality of other dictionaries,
One or more codes obtained as a result of the search and a dictionary reference device for outputting the priorities thereof are provided.

That is, specifically, a plurality of input devices for inputting a pattern, a dictionary provided for each of the input devices, and a dictionary for storing one or more codes corresponding to the patterns together with a priority order, Receiving an input pattern input through a certain input device, a dictionary corresponding to the input device and a pattern registered in a plurality of other dictionaries that match or most similar to the input pattern are selected. A dictionary reference device for searching and outputting the code and the priority, the dictionary reference device being configured to hold a dictionary priority which is a priority set in advance for a plurality of dictionaries. is there.

A third feature of the present invention is that a plurality of input devices for at least receiving a given pattern,
In a pattern information processing apparatus having a dictionary for storing the input pattern and one or more codes corresponding to the input pattern together with the priority order for each of the input devices, when any of the plurality of dictionaries is updated, There is provided a dictionary editing device for registering a relationship between an input pattern, which is an update content, and a code corresponding thereto in another dictionary.

[0017]

According to the first feature of the present invention, first, an input pattern such as a character, a voice, a code string, etc., inputted through the input device is transmitted to the recognition device together with the belonging attribute designated by the attribute designation device. Can be Next, the recognition device receives the input pattern and the belonging attribute, and searches for a pattern having the corresponding belonging attribute and which is most similar to the input pattern from a plurality of patterns existing in the dictionary. Outputs the category to which belongs. The category is a superordinate concept of the pattern, and one category has a relationship to which at least one or more patterns belong. Alternatively, when the input pattern is a code string, the input pattern is searched from a plurality of patterns existing in the dictionary, and a code indicating a process associated with the pattern is output. Further, the dictionary editing device examines the contents of the dictionary, extracts a pattern that will be commonly used in a certain group, sets the belonging attribute of the pattern to the group, and performs pattern registration processing.

By performing such processing, the user belonging to the group can perform pattern recognition using the pattern without registering the pattern by himself.
Pattern processing can be performed.

According to a second feature of the present invention, first,
An input pattern such as a character, a voice, and a code string input through the input device is sent to the dictionary reference device. Next, the dictionary reference device receives the input pattern, and matches the input pattern from a dictionary corresponding to the input device into which the input pattern is input and a plurality of patterns existing in other dictionaries. Alternatively, a search is made for the most similar one, and the code is re-prioritized and output based on dictionary priorities, which are priorities set in advance for a plurality of dictionaries.

As described above, by performing processing by referring to the dictionary registration contents of another user, pattern recognition and pattern processing using the pattern can be performed without registering the pattern by itself. . In addition, by assigning priorities based on dictionary priorities, even if different codes are defined for the same pattern between your own dictionary and other user's dictionaries, the definition of your own dictionary is given priority. Becomes possible.

According to a third feature of the present invention, the dictionary editing device examines the contents of a plurality of dictionaries, and updates information (input patterns, codes corresponding thereto, and priorities) in a certain dictionary. If it is detected that the updated input pattern has not been defined, registration processing of the updated information is performed for another dictionary that does not define the updated input pattern.

By performing such processing, when a certain user updates the contents of the dictionary, other users can use the updated information without registering the contents themselves. .

[0023]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Here, as an embodiment according to the present invention, an input system for handwritten characters using an online character pattern recognition process will be described as an example.

Of course, in the present invention, the patterns are not only characters, but also various patterns such as voices and graphics, and
Needless to say, it can be applied to the code string pattern.

First, regarding the system configuration of this embodiment,
This will be described with reference to FIG.

The present system includes an input device 1, a dictionary 2, a main memory 3, a recognition device 4, and a dictionary editing device 5.

The input device includes a plurality of input units (1-
1, 1-2, 1-3). Furthermore, the main memory 3 has a configuration in which an attribute designation device 3-1 is realized.

The dictionary 2 includes a plurality of dictionaries (2-0, 2-1 and 2-2) in consideration of categories and attributes as described later.
-2, 2-3, 2-4).

Furthermore, the dictionary editing device 5 has a configuration including a learning device 5-1 and a dictionary optimizing device 5-2.

Here, the components of the present system will be described.

First, the input device 1 includes a pen, a liquid crystal display device 1-1-1, a liquid crystal display device controller 1-1-2, a tablet 1-1-3, and a tablet controller 1
-1-4 and three input units.

The user writes a stroke on the tablet using a pen to input a character pattern,
The user name is specified, a character pattern to be learned to be recognized is input, and a command to learn is input.

As described above, the strokes written by the user using the pen on the tablet correspond to the liquid crystal display device 1-1.
The display process is performed by the liquid crystal display device controller 1-1-2. The liquid crystal display controller 1-1-2 includes, for example, a CPU, a ROM, a RAM, various CMOSs,
This is realized by a program or the like stored in the OM. The input device also has a function of outputting and displaying necessary information.

The display device is not limited to a device using a liquid crystal element, but may be a display device including other elements, for example, an EL element.

The user can use the input device with the same feeling as using paper for writing characters.

As described above, in the system configuration example of this embodiment, a plurality of input units (1-1, 1
-2, 1-3) to constitute the input device.

The input device 1 transmits input information such as a character pattern provided by the user and ID information added to the tablet to which the input information has been input to a signal line 100.
Output to the top. Note that the ID information added to the tablet refers to information set in advance to identify the input unit. For example, different alphanumeric characters and the like are individually set, and the ID information is set on the signal line 100. The output circuit for outputting the data to the device may be realized by an electronic device such as various CMOSs.

The dictionary 2 defines attributes such as patterns, categories to which the patterns belong, character types, and affiliations.
Various types of data are stored in the dictionary and are classified according to attributes in order to improve processing efficiency at the time of recognition processing described later.

Here, the category is a general concept of the pattern, and at least one pattern belongs to one category.

The attributes include the name of the group to which the user belongs.

In this embodiment, as shown in FIG.
0: Everyone, "2-1: User 1", "2-2: User 2", "2-3: User 3", "2-4: Group 1" (In this embodiment, this system is used. The members of the group to be created are user 1, user 2, user 3
), And the data in the dictionary is classified according to the belonging attribute (data belonging to the user). However, the data may be classified according to another attribute such as a character type attribute.

The attribute specifying device 3-1 is provided in the main memory 3 and has a function of specifying an attribute at the time of pattern recognition processing.

The main memory 3 is, for example, a RAM
And the like, or an optical / magnetic disk device.

Further, the attribute designation device 3-1 is connected to the signal line 1
When a user name and the ID information of the tablet currently used by the user are received via the “00”, the information processing apparatus also has a function of storing the information. In addition, when a certain application is activated and receives a character type or the like designated by the application, a process of retaining the character type is also performed.

Then, thereafter, the input to the input device 1
When a signal and ID information of the tablet are received via the signal line 100, attributes such as affiliation and character type corresponding to the information are output on the signal line 100.

When the recognition device 4 receives a pattern to be recognized and attributes such as affiliation and character type via the signal line 100,
By referring to the contents of the dictionary 2 via the signal line 101, a pattern having the attribute such as affiliation or character type is searched for a pattern most similar to the shape of the pattern to be recognized. Output on line 100.

For example, it is realized by an electronic device such as a CPU, a ROM, and a RAM, and a program stored in the ROM.

At this time, whether or not the pattern shapes are similar is evaluated by using a predetermined matching function or the like.
For example, it is described in JP-A-63-254589 and JP-A-2-53193.

According to this, for example, the "pattern shape" may be represented by coordinate points obtained by approximating each of the strokes (the strokes represent points and lines constituting Chinese characters) with N points. Further, the “matching function” may be expressed by a linear expression in which each coordinate point in the input pattern and the pattern in the dictionary is used as a parameter.

As described above, the dictionary editing device 5 has the learning device 5-1 and the dictionary optimizing device 5-2. The dictionary editing device 5 includes, for example, a CPU, ROM, RA
It is realized by an electronic device such as M and a program stored in a ROM.

The learning device 5-1 outputs a pattern to be learned so as to be recognized, a command to learn, an attribute of the pattern such as a character type, a category, etc., output from the input device 1; When the output belonging attribute is received via the signal line 100, the pattern information is registered in a dictionary to perform so-called learning. That is, the dictionary pattern of the user
A process for newly registering this pattern information is performed.

The learning device 5-1 inputs pattern information by a user's operation, and automatically fetches the pattern information for erroneous recognition in character recognition processing of a character pattern given by each user and registers the pattern information in a dictionary. It is preferable from the viewpoint of operability that such a configuration is adopted.

Here, the erroneous recognition is, for example, when the user selects the N-th candidate from among the candidate characters as a result of recognition of the input pattern, the first candidate is erroneous recognition. As described above, it is preferable that the input pattern is registered in the dictionary as the N-th character code.

The dictionary optimizing device 5-2 is a main part of the present invention.

The dictionary optimizing device 5-2 performs a process of detecting an excellent portion of a certain user dictionary and making the portion available to all users who are members of the group. This processing will be described later in detail.

Here, the “excellent part” means, for example, a case where a pattern commonly used by members of a group, such as a symbol of a group, is registered in a user dictionary of a certain member. This refers to the registration information.

As described above, in this embodiment, the description is made by taking the character recognition system as an example. However, a general pattern matching system such as voice pattern recognition and kana-kanji conversion is also used in a dictionary optimization system. By providing and operating 5-2, it is possible to perform pattern learning of various patterns for a multi-user. Next, an operation example of the present system will be described.

FIG. 2 shows a state in which the user 1 and the user 2 have registered the “abbreviated character” of the character “kan” by the processing performed by the learning device 5-1.

In the dictionaries 2-1 and 2-2, abbreviation character patterns of the kanji "mon" are registered in the fonts written by the respective users. That is, a common pattern is registered in two user dictionaries, which is a majority of three members of a certain group 1.

In general, empirical facts show that pattern information that is used by a majority of people and is convenient is often useful to other people.

In FIG. 3, as a result of the processing performed by the dictionary optimizing device 5-2, this pattern information is detected as an excellent part in these user dictionaries (2-1, 2-2). , Registered in the dictionary 2-4 of the group 1,
This shows a state where these (abbreviated character patterns of the kanji “mon”) have been deleted from the two user dictionaries (2-1, 2-2).

Of course, two user dictionaries (2-1, 2-1)
From 2-2), the common pattern information may be registered without deleting it.

Next, the configuration of the dictionary optimization device 5-2 will be described in more detail with reference to an embodiment.

First, the first embodiment will be described with reference to FIGS.

FIG. 4 shows a hardware configuration example, and FIG. 5 shows a processing procedure.

In this embodiment, the dictionary optimizing device 5-2
Is a dictionary update detection device 11, a group extraction device 12, a user extraction device 13 in a group, and a dictionary information reference device 1.
4. Same pattern information counter 15, dictionary update device 16
And a signal line connecting them.
Each component is, for example, CPU, ROM, RA
M, various CMOSs, a program stored in a ROM, and the like.

When the pattern information in the dictionary 2 has been updated, the dictionary update detecting device 11 detects the pattern information which is the update content and outputs the information on the signal line 200.

For example, a configuration is provided in which a signal line is provided between the learning device 5-1 and the learning device 5-1 receives the updated content when the content of the dictionary 2 is updated.

Alternatively, the update time of the dictionary file is checked for each unit time, and the updated time is compared with the last update time.
If later, it is determined that the update process has been newly performed,
A comparison process with the contents of the old dictionary is performed to obtain difference information. Such difference information may be stored in the main memory 3, for example. Then, a process of recording the updated dictionary content as the old dictionary content is performed in order to detect whether or not the next update process has occurred.

When receiving the pattern information from the signal line 200, the group extracting device 12 examines the belonging attribute of the updated pattern information, that is, the user, and extracts all the groups to which the user belongs. Information on the signal line 201 is output. This is because a user may belong to a plurality of groups. If the user is allowed to belong to a plurality of groups, it is preferable to prepare a plurality of group dictionaries.

For this purpose, the correspondence between each user and the group must be stored in advance in, for example, the main memory 3. Of course, a storage unit for storing such correspondence may be provided in the group extraction device 12, and if this system is configured on a workstation system, the function of the workstation system is used. Is also good. That is, some OSs (Operating Systems) installed on the workstation have a function of defining the relationship between the user and the group, and this function may be used.

When receiving the group information via the signal line 201, the in-group user extracting device 13 extracts all the users belonging to the group, and
02. For this purpose, it is necessary to store the correspondence between each user and the group in advance, and a storage unit for that purpose is required. This storage section is provided in the main memory 3, in the configuration provided in the in-group user extraction device 13 itself, or in the workstation system if the present system is configured on a workstation system. Configuration and the like are conceivable.

The dictionary information reference device 14 receives the pattern information via the signal line 200 and, when receiving the user information via the signal line 202, reads the contents of the dictionary belonging to the user via the signal line 101. refer. Then, if the pattern information and the category received through the signal line 200 are the same and the shapes of the patterns are similar, a signal is output on the signal line 203. At this time, the determination as to whether or not the shapes of the patterns are similar is evaluated by a matching function used in the recognition device 4,
The evaluation is performed by comparing the evaluation value with a predetermined threshold value.

For example, when the evaluation value becomes larger than the threshold value, processing such as judging that the pattern shapes are similar is performed.

The same pattern information counter 15 receives pattern information having similar pattern shapes via the signal line 203 and counts the number of similar patterns.

When the count value satisfies a predetermined condition, the received information is output to a signal line 204.

For example, the received information may be output to the signal line 204 on condition that the count value exceeds a certain number N (N is a natural number), such as a count value of 2 or more. In this case, the fixed number N is a function of the number of members in the group (for example, “N = (number of members) × A + B”).
(However, A and B are assumed to be certain real numbers). The dictionary updating device 16 receives the pattern information of the user belonging to the same group and the information on the designation of the group via the signal line 204, and transfers the representative pattern information to the user dictionary belonging to the group via the signal line 101. Perform the registration process.

At this time, the pattern shape of the representative pattern information may be the average of the pattern shapes of all the pattern information received via the signal line 203.

When the averaging process cannot be performed because the pattern shape is not represented by a numerical value (for example, when a character pattern structure analysis process is performed), one or more of them may be set as the representative pattern shape.

The dictionary updating device 16 is connected to the signal line 204
The processing for deleting the pattern information received via the. Of course, without performing such deletion processing, the signal line 20
4 may be left as it is.

FIG. 5 shows an operation procedure of the dictionary optimizing device 5-2 having the above configuration.

First, in step 10, the dictionary update detecting device 11 operates to determine whether or not the dictionary has been updated.
If the contents of the dictionary have been updated, the process branches to step 20, and the following processing is performed.

At step 20, the group extracting device 12 operates to list the groups to which the dictionary user belongs, and at step 30, the following processing is performed on those groups.

That is, in step 40, the in-group user extraction device 13 operates to list the user dictionaries of the users belonging to the group.

Then, in step 50, the following processing is performed on these user dictionaries. That is, in step 60, the dictionary information reference device 14 operates to list the same pattern information as the update content, and further, in step 70, the same pattern information counter 15 operates to execute the same pattern information as the update content. The pattern information is counted. Next, when the counted value satisfies a predetermined condition (for example, whether or not a value exceeds a predetermined threshold is a determination condition, and the value exceeds the threshold),
In step 80, the dictionary updating device 16 operates,
The same pattern information as the updated content is registered in the user dictionary owned by the group, and in step 90, a process of deleting the same pattern information as the updated content from each user dictionary is performed.

The above processing is repeatedly performed.
It is also preferable that a meta group located one level higher than the group ("meta" means one level higher) can be defined, and the dictionary can be updated by similar processing.

As described above, according to this embodiment, an apparatus for recognizing various patterns intended for a multi-user can be constructed. When a plurality of users register the same or similar patterns, the other users can register the same or similar patterns. An apparatus can be provided that allows the pattern to be used.

Next, a second embodiment according to the present invention will be described.

This embodiment is an example of the configuration of the dictionary optimization device 5-2 to which a so-called genetic algorithm is applied. The entire system configuration is as shown in FIG. 1. FIG. 6 shows a detailed hardware configuration of the dictionary optimizing device 5-2 in FIG. 1, and FIG. 7 shows a processing procedure.

The dictionary optimizing device 5-2 in the present embodiment
Is composed of a fitness memory 21, a parent pattern selection device 22, a child pattern generation device 23, an operation determination device 24, a child pattern memory 25, a generation alternation device 26, and a signal line connecting these. Each component is realized by, for example, a CPU, a ROM, a RAM, a CMOS, software stored in the ROM, and the like.

Here, the function of each component will be described.

The fitness memory 21 stores the fitness fi of each pattern information i registered in the dictionary 2.

The “fitness fi” shows a higher value as it is frequently used for character recognition, and shows a similarity in a processing result using a matching evaluation function, which is a function of the recognition device 4. The higher the index, "similarity", the higher the value, and the wider the range of belonging attributes (for example, the fact that belonging to a group is used by many users rather than belonging to an individual It is a kind of parameter that indicates a high value. In addition, if other conditions are the same, generally, a pattern belonging to a group has a higher fitness fi than belonging to a single user than belonging to a single user. In addition, if the pattern belongs to a plurality of groups, the fitness of a pattern belonging to a larger group increases.

Further, when a character pattern belonging to another category is input, if it is erroneously recognized as the first candidate, the fitness is determined so that the value becomes low.

A restriction condition may be given to the fitness value, for example, by setting the sum of the fitnesses fij of the same category j to be a constant value.

The fitness fi is preferably stored in the dictionary 2 shown in FIG. 1 as one of the attributes of the pattern information.

The generation alternation device 26 determines the timing and updates the contents of the dictionary at this timing. The timing is, for example, at regular time intervals or when the contents of the dictionary are watched, and when a certain condition is satisfied, such as when the accumulation of dictionary updates by the user from the previous timing reaches a certain amount.

The name of the name of the generation alternation is obtained by replacing n patterns registered in the dictionary 2 with another m patterns when updating the contents of the dictionary.
This is because generational change occurs from the generation of parents to the generation of m children.

Hereinafter, the pattern before replacement is called a parent pattern, and the pattern after replacement is called a child pattern.

The generation alternation device 26 outputs the signal of the generation alternation on the signal line 304 and, after the child pattern is generated in the child pattern memory 25, deletes the parent pattern from the dictionary via the signal line 101. Then, the content of the dictionary is updated by performing a process of registering a child pattern.

Next, the parent pattern selecting device 2 shown in FIG.
2 receives the designation of the category j via the signal line 304, and further receives the signal line 301 from the child pattern generation device 23.
The number is specified via. Then, referring to the contents of the dictionary 2 shown in FIG. 1, the designated number of pieces of pattern information are selected from the pattern information having the attribute of the category j.

At this time, the probability that the pattern information ij is selected follows the formula fi / Σfij according to the fitness fi (where Σ means summing up all i).

The child pattern generation device 23 includes a signal line 301
The timing information of the generation change is received, and further, the operation including copy, crossover, and mutation, and the number n of parameters are received via the signal line 302. Then, the number of parameters is output on the signal line 301, and parent pattern information is received from the parent pattern selection device 22 by the number n in accordance with the output signal.

Then, the operation is performed on the parent pattern information, and new child pattern information including the fitness is generated.
The signal is output on a signal line 303. An operation copy is 1
That is, one parent pattern is copied to generate one child pattern. Crossover is to generate one child pattern by transferring partial features of two parent patterns.
Mutation is the generation of a child pattern that differs from the characteristics of the parent pattern. These operations will be described later in detail.

Further, the operation determining device 24 transmits information on the number of operation commands and parameters to the signal line 3.
02.

Although the number of parameters is determined by the operation, the probability of outputting the operation k follows a predetermined probability.

The child pattern memory 25 stores new pattern information including the fitness, received via the signal line 303.

FIG. 7 shows a flow of operation in the above-described apparatus.

First, in step 110, the generation alternation device 26 notifies the timing of the generation alternation when a predetermined time has elapsed. At this timing, when the apparatus detects that the user has changed the situation, such as the registration of a new character in the dictionary, the recognition process for the character belonging to a certain category is performed a predetermined number of times or more. It is also preferable to set the time.

The following processing is executed for each category.

In step 120, the following process is repeated N times to generate N child patterns.

In step 130, the operation determining device 24 determines (1) “copy”, (2) “crossover”, and (3)
One operation is selected and determined from the operations such as “mutation”. At this time, which one is selected
For example, in the order of (1) “copy”, (2) “crossover”, and (3) “mutation”, the probabilities are determined, for example, 0.5, 0.495, 0.005, and the probabilities are determined.

Then, in step 140, the process of selecting the parent pattern information is performed, and depending on the type of the operation,
For example, the numbers to be selected as 1, 2, and 0 are different in the order of (1) “copy”, (2) “crossover”, and (3) “mutation”.

As a rule for determining which pattern is to be selected as a parent pattern, the higher the pattern fitness is, the higher the selection probability is.

Next, in step 150, child pattern information is generated based on the operation and the parent pattern.

The meaning of the operations "copy", "crossover" and "mutation" will be described.

(1) "Copy" is an operation for copying the parent pattern information as it is to generate child pattern information.

(2) "Crossover" is an operation in which, for example, two parts are extracted from each of two pieces of parent pattern information and combined to generate a pattern, which is used as child pattern information.

At this time, it is preferable that the intersecting position is a break between radicals if it is a kanji character, and a break between strokes of a pattern if it is not a kanji character.

(3) “Mutation” is an operation for performing processing such as replacing part of information of a parent pattern for newly generating character information with part of pattern information belonging to another category. is there.

After generating the N child patterns, in step 160, the parent pattern information and the child pattern information are exchanged. Generally, it is preferable that the parent pattern information and the child pattern information have the same number.

Next, how to give the fitness will be described with reference to FIG.
This will be described with reference to FIG.

The fitness is determined in step 21 in FIG.
0, update in cases such as shown in step 230 and step 260. Here, it is assumed that the fitness is given for each image ("image" is referred to as "shadow" and means a point or a line for assembling characters) constituting one pattern.
A case will be described in which the fitness of the entire pattern composed of a picture is determined based on the fitness given to the picture.

If the user has registered a new pattern, in step 211, the fitness f
i = 0.5.

In the case where the belonging attribute of the pattern is changed, and particularly in this case, when the belonging attribute is expanded, the step 212 is performed so that the fitness is increased.
In the equation, fi = min (1, fi + 0.25) (where min (a, b) means to select a smaller one of a and b) to give the fitness.

If the belonging attribute is reduced, fi = max (0, fi−0.25) (where max (a, b) is a , B). Of course, the processing in step 212 or step 213 may not be performed.

If the input pattern is recognized in step 230, the fitness of each stroke of the character that is the first candidate is determined in step 240 by fi = min (1, fi + 0.25). ), And in step 250, the input pattern is a character belonging to the same category as the first character.
The fitness of each stroke of the candidate character is given.

In step 251, all characters a (a:
Fi = min (1, representing an arbitrary character) if it meets the criterion shown in step 252, that is, if there is an image whose distance value to the input pattern is smaller than the character as the first candidate. fi + 0.25).

Here, the “distance value” is a function value of a matching function calculated between the input pattern and the dictionary pattern in the dictionary. For example, when the pattern shape is represented by approximate point coordinate values obtained by approximating N points and the matching function is represented by the sum of the differences between the approximate points coordinates of the input pattern and the dictionary pattern, the sum is the distance value.

However, if the candidate is the first candidate in step 260 but is treated as an erroneous recognition, then in step 270, a fitness degree of fi = max (0, fi-0.25) is given.

Here, in the case of a kanji pattern, it is preferable to give the fitness by using a sub-pattern such as a radical as one unit, rather than giving the fitness to each stroke, since the processing is simpler and is therefore preferable. .

Now, FIGS. 9 and 10 show the category “A”.
An example of updating a dictionary pattern (category for the character pattern “A”) will be described.

FIG. 10 exemplifies a child pattern generated by operation copying, crossover, and mutation based on the parent pattern shown in FIG.

As shown in FIG. 9, when several patterns of characters “A” are registered in the dictionary (from p1 of pattern 1 to p7 of pattern 7), the generation At the time of switching, if the operation of “copy” is selected with a probability of 0.5 in the operation determination, the probability of (fitness of p1 / fitness of Σpi) (Σ means the sum of i) Then, p1 is selected, and the p child 1 shown in FIG. 9 is generated.

Here, the fitness of pi is obtained from the fitness given to each image. For example, it may be obtained by processing such as taking the average value of the fitness given to each image.

Therefore, for example, the fitness of p1 is “1”, “1”, and “1” for each picture as shown in FIG. ing.

Therefore, for example, the fitness of p1 is: fitness of p1 = (1 + 1 + 1) / 3 = 1.

The same processing is applied to p2 to p7 so that the conformity is “0.5”, “1”, “1”, “0.5”, “0.5”.
5 "and" 0.25 ".

Therefore, the probability that p1 is selected is as follows: Probability that p1 is selected = 1 / (1 + 0.5 + 1 + 1 + 0.5)
+ 0.5 + 0.25) ≒ 0.266.

Similarly, with a probability corresponding to the fitness,
An operator and a parent pattern are selected, and p children 2 and p children 3 are generated by the operations such as “crossover” and “mutation”, as shown in FIG.

In the processing described in the first and second embodiments, not only characters but also figures,
This can be performed as a pattern of various information such as voice.

If the dictionary 2 is classified into categories such as character types, the operation of the dictionary optimizing device 5-2 can be changed for each category. In other words, a dictionary that has a high degree of commonality among users is actively subjected to dictionary optimization, and the other dictionary is not optimized. For example,
In character recognition, hiragana dictionary optimization may be performed, and other character types may not be performed.

Also, in the case where the input pattern is a code string, such as Kana-Kanji conversion, (1) replace the recognition device 4 with the Kana-Kanji conversion device 4, and (2) convert the registered contents of the dictionary 2 into the conversion code for the input code sequence. Replace with column definition,
(3) The present invention can be applied in the same manner except that the input device 1 is replaced by another device, ie, the input device 1 and the display device 6.

The recognition device 4 shown in FIG. 1 is changed to a kana-kanji conversion device 4, that is, when an input code string and an attribute such as affiliation are received via a signal line 100, the contents of the dictionary 2 are received via a signal line 101. , The input code string is searched from the dictionary having the attribute such as the affiliation, and the corresponding conversion code string is output on the signal line 100.

The dictionary optimizing device 5-2 is a main part of the present invention.

The dictionary optimizing device 5-2 detects a superior portion of a certain user dictionary and performs a process of making the portion available to all users who are members of the group. This processing can be realized by the configuration of FIG. 4 and the operation of FIG. 5, as in the first embodiment.

Here, the “excellent part” is, for example, a kana-kanji conversion pattern commonly used by members of the group because it is a rule in the group, and is registered in the user dictionary of the member. Means the registration information at the time of registration.

For example, if there is a common kana-kanji conversion pattern in the registered information of the user dictionary, this is detected as an excellent part, and this part is made available to all users who are members of the group.

FIGS. 11 and 12 show an example of processing in the case of the kana-kanji conversion device. As in the first embodiment, “2-0: all”, “2-1: user 1”,
"2-2: User 2", "2-3: User 3", "2
-4: group 1 "(in the present embodiment, the members of the group using this system are user 1, user 2 and user 3) and the data in the dictionary are classified. Of course, they may be classified according to other attributes such as.

FIG. 11 shows a state where two users have individually registered the same fixed phrase by the learning device 5-1.

In FIG. 12, the dictionary optimizing device 5-2 operates from the state of FIG. 11, and it is the rule in the group that the kana-kanji conversion of the character code string “Bishamon” is performed as “Bishamon”. Is registered in a group dictionary that can be used by all users in the group, and is deleted from the two user dictionaries. Alternatively, the same content may be registered in the user dictionaries of all the members in the group.

In the above, the embodiment in the case where the input pattern is a code string has been described by taking Kana-Kanji conversion as an example. In other cases, the pattern and its processing are registered in a dictionary, and based on this, The present invention can be similarly applied to the definition of the system operation. That is, when two or more persons in the group rewrite the user dictionary and customize the system operation to improve the operability, all the members in the group can obtain the customized operability.

Further, unlike the case of the first embodiment,
A system configuration in which a plurality of handwriting input devices in which the input device 1, the dictionary 2, the main memory 3, and the recognition device 4 are integrated are connected via a network may be adopted.

Although the following embodiments will be described by taking Kana-Kanji conversion as an example, the present invention can be applied to a pattern recognition device in the same manner except for the above (1), (2) and (3). Further, the present invention can be similarly applied to the definition of the system operation. As a third embodiment,
In particular, a method of referring to a dictionary of another user without providing a group dictionary will be described.

In FIG. 13, the word processor is the input device 1
(Input device device 1-1 and input device controller 1-2), display device 6 (display device device 6-1 and display device controller 6-2), main memory 3, kana-kan conversion device 4 (processing device 4-1). And dictionary reference device 4-
2) Dictionary 2 Some word processors are provided with a learning device 5-1 so that a user can register a fixed phrase used by himself / herself.

The input device 1 for receiving a pattern as an input code string is generally a keyboard, but may use a method of selecting a code menu displayed on the display device 6 with a mouse. The pattern recognition device as described in the embodiment may be used as the input device 1 to input a code string in which a handwritten character is recognized (in this case, there is no need to provide the display device 6 separately). May be replaced with a device capable of inputting a.

In the dictionary 2, a plurality of input code strings and a plurality of conversion code strings corresponding to the input code strings are defined together with their priorities (see dictionary 2 (j) in FIG. 14). In the system configuration of the first embodiment, an attribute indicating the owner of the dictionary is also defined in the dictionary 2. However, if there is a word processor for each user as in the system configuration of FIG. If processing is performed using an ID, the dictionary 2 need not have an attribute indicating the owner of the dictionary.

The kana-kan conversion device 4 includes a processing device 4-1 and a dictionary reference device 4-2. The processing device 4-1 includes:
The input code string input by the user through the input device 1 is received via the main memory 3, and one word is cut out from the input code string and passed to the dictionary reference device 4-2.

The dictionary reference device 4-2 searches the dictionary 2 for the input code string for one word. Dictionary 2 for each word processor
Is connected via a network, and the dictionary reference device 4-
2 searches for the input code string including the dictionary of the word processor of another user. In the case where a plurality of conversion code strings are defined for the input code string, the priority is determined and the conversion code string is passed to the processing device 4-1.

FIG. 14 shows the operation procedure of the dictionary reference device 4-2. First, in step 401, a candidate number counter c is initialized to zero. Then, in step 402, the following processing is repeated for all the dictionaries of the users belonging to the same group. For this purpose, it is preferable to have a dictionary priority table as shown in FIG. At this time,
If a conversion code for an input code string is defined in a dictionary having a high dictionary priority, the processing may be terminated without performing the processing for all the dictionaries of the users belonging to the same group.

In the figure, as an example of a dictionary of a user belonging to the same group, a definition part of an input code string “Bishamon” is shown for the dictionary 2 (i) in the word processor i and the dictionary 2 (j) in the word processor j. 2 (i)
Not defined, and two conversion codes are defined in 2 (j).

At step 403, it is checked whether or not the input code string is defined in the dictionary 2, and if it is defined, steps 405 and 406 are performed for all defined conversion codes after step 404. That is, the priority order for the input code sequence is calculated, the priority order and the conversion code are registered in the candidate table (c), and c is updated. As the priority order, a dictionary priority is determined in advance as shown in the example in the figure, and the sum with the priority order of the conversion code is calculated and used. Alternatively, the calculation may be performed using a product or using another priority function.

At step 407, the c candidates registered in the candidate table are sorted by priority and output at step 408. That is, the candidate table sorted by priority is passed to the processing device 4-1. Although the figure shows an example of the candidate table, the user of the dictionary 2 (i) can obtain the conversion code even though it is not defined in his own dictionary.

The processing device 4-1 sets the highest priority as a Kana-Kanji conversion result.

The display device 6 displays an input code string input by the user using the input device 1, and after converting this code string into a Kana-Kanji string by the dictionary reference device 4, displays a character code string after the Kana-Kanji conversion. .

The learning device 5-1 performs learning on kana-kanji conversion for each user and compiles the dictionary 2. That is, the learning function for the priority of the code string after the kana-kanji conversion based on the fixed phrase registration and the history of the kana-kanji conversion is performed.

Next, as a fourth embodiment, a group dictionary is provided in the same manner as in the first embodiment.
A case will be described in which the dictionary 2, the main memory 3, and the kana-kan conversion device 4 (the recognition device 4 in the case of a handwriting input device) are integrated.

In the system configuration in this case, the following configuration is added to the system configuration of the second embodiment. That is,
The dictionary optimization device 5-2 and the group dictionary 2 (k) are added. Dictionary optimization device 5-2 and group dictionary 2 (k)
Is the same as that described in the first embodiment. The processing procedure of the dictionary optimization device 5-2 follows the flow of FIG. 5 as in the first embodiment.

Next, as a fifth embodiment, a method of distributing the contents of one dictionary to another dictionary without providing a group dictionary will be described.

FIG. 15 shows the system configuration. In the word processor described in the second embodiment of FIG. 13, the dictionary reference device 4-2 is devised in order to refer to the dictionary of another user. Or a general kana-kan conversion device 4 which only refers to the dictionary.

A dictionary optimizing device 5-2 is particularly provided for distributing the contents of one dictionary to another dictionary. FIG. 16 shows an operation flow of the dictionary optimization device 5-2.

First, in step 10, it is checked whether the dictionary has been updated, and if any dictionary has been updated, the following processing is performed. The list of the groups to which the users of the dictionary updated in step 20 are listed, and the following processing is performed for all the groups. At step 40, the user dictionaries of the users belonging to the group are listed, and at step 50, step 60 is performed for all the user dictionaries. That is, the same information (conversion code for the input code string) as the update content is registered. However, since the conversion code may already be registered for the input code string, the priority of the conversion code for the input code string is set, and the priority of the conversion code of the updated content is the last.

Next, as a sixth embodiment, a method will be described in which the dictionary optimizing device 5-2 is provided in a word processor and the contents of a certain dictionary are copied to its own dictionary.

FIG. 17 shows the system configuration. The dictionary optimizing device 5-2 described in the fifth embodiment of FIG. 15 is in a word processor. FIG. 18 shows an operation flow of the dictionary optimization device 5-2.

First, in step 501, the following is repeated for all the dictionaries of other users registered in the dictionary priority table. At step 502, it is checked whether the dictionary has been updated. If the dictionary has been updated, the following processing is performed. In step 503, it is checked whether or not the input code has already been defined in the dictionary 2 with respect to the input code. If not, the update content of the dictionary of another user is registered in the dictionary in step 504. . The priorities of the conversion codes at this time are the same as those described in the third embodiment.

The figure shows the contents of the dictionary 2 of each user before and after the operation of the dictionary optimizing device 5-2. Regardless, a conversion code can be obtained.

[0177]

According to the present invention, a dictionary used for various pattern recognition, kana-kanji conversion, system operation, etc.
The learning can be advanced by a plurality of users, and can be performed very efficiently and effectively as compared with the case where each of the users advances individually.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a system according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram of learning using a dictionary.

FIG. 3 is an explanatory diagram of learning using a dictionary.

FIG. 4 is a diagram illustrating a configuration example of a learning device 5-1.

FIG. 5 is an explanatory diagram of a processing procedure according to the present invention.

FIG. 6 is a diagram illustrating a configuration example of a learning device 5-1 according to another embodiment.

FIG. 7 is an explanatory diagram of a processing procedure in another embodiment.

FIG. 8 is an explanatory diagram of a procedure for giving a fitness fi.

FIG. 9 is an explanatory diagram of learning using a dictionary in another embodiment.

FIG. 10 is an explanatory diagram of learning using a dictionary according to another embodiment.

FIG. 11 is an explanatory diagram of learning using a dictionary.

FIG. 12 is an explanatory diagram of learning using a dictionary.

FIG. 13 is a system configuration diagram according to the present invention.

FIG. 14 is an explanatory diagram of a processing procedure according to the present invention.

FIG. 15 is a system configuration diagram according to the present invention.

FIG. 16 is an explanatory diagram of a processing procedure according to the present invention.

FIG. 17 is a system configuration diagram according to the present invention.

FIG. 18 is an explanatory diagram of a processing procedure according to the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Input device, 2 ... Dictionary, 3 ... Main memory, 3-1 ...
Attribute designation device, 4 ... recognition device, 5 ... dictionary editing device, 5-
1. Learning device, 5-2 Dictionary optimization device.

────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Toshifumi Arai 7-1-1, Omika-cho, Hitachi City, Ibaraki Prefecture Within Hitachi Research Laboratory, Hitachi, Ltd. (72) Inventor Masaki Miura 7-1-1, Omika-cho, Hitachi City, Ibaraki Prefecture No. 1 Hitachi, Ltd., Hitachi Laboratory (72) Inventor Keiko Gunji 1-1-1, Omika-cho, Hitachi City, Ibaraki Prefecture Hitachi, Ltd., Hitachi Research Laboratory (72) Inventor Yasushi Fukunaga 7, Omika-cho, Hitachi City, Ibaraki Prefecture No. 1-1 Hitachi Research Laboratory, Hitachi Research Laboratory (56) References JP-A-6-342473 (JP, A) JP-A-58-223887 (JP, A) (58) Fields investigated (Int. Cl. ⁷⁾ G06K 9/68 G10L 15/06 JICST file (JOIS)

Claims (7)

(57) [Claims] 1. A plurality of input devices for at least receiving a given pattern, and a plurality of storage units classified into a plurality corresponding to a plurality of users and storing a pattern given via the input device. And a processing unit, wherein at least one of the storage units is a shared storage unit that can be sequentially updated by the processing unit, and the processing unit is at least two different storage units. when the pattern of each approximating the storage unit is stored, the pattern information processing apparatus characterized by storing the pattern of any one of approximation before SL shared storage unit. 2. A pattern information processing system comprising: a plurality of input devices for receiving at least a given pattern; and a plurality of dictionaries having a configuration in which data is classified into a plurality and registered in correspondence with a plurality of users. In the apparatus, each of the plurality of dictionaries is registered in advance with an attribute belonging to each user.
At least one dictionary is referred to by the plurality of users.
A irradiation capable shared dictionary, before on the basis of the membership attribute registered in each of the dictionary
A dictionary editing device that edits the contents of each of the dictionaries. When the pattern given via the input device satisfies a predetermined condition, the dictionary editing device performs an editing process of data in the dictionary. Then, the pattern and the code corresponding to the pattern are registered in the common dictionary, and are registered in the common dictionary.
Referred to by the user having the same affiliation attributes and Attributes
Pattern information processing apparatus, wherein said a shared dictionary Exiled following updatable be. 3. The pattern information processing apparatus according to claim 2, wherein the belonging attribute of each user is a name of a group including a plurality of users as members. 4. The pattern according to claim 2, wherein the data editing process performed by the dictionary editing device is an addition or deletion of a pattern registered in the dictionary. Information processing device. 5. The pattern editing apparatus according to claim 2, wherein information obtained by adding a pattern belonging attribute to the pattern registered in the dictionary is registered in the dictionary, and the dictionary editing device updates the pattern belonging attribute. A pattern information processing apparatus characterized by the following. 6. A plurality of input devices for receiving at least a given pattern, and a dictionary in which the input patterns and one or more codes corresponding to the input patterns are stored together with a priority order for each of the input devices. In the pattern information processing apparatus, a code corresponding to a pattern given via an input device is searched from a dictionary corresponding to the input device and a plurality of other dictionaries, and one or more codes obtained as a result of the search are obtained. Print code and its priority, or search results
One or more codes in the order of priority.
A pattern information processing apparatus, comprising: a dictionary reference device for inputting data. 7. The dictionary reference device according to claim 6, wherein the dictionary reference device holds dictionary priorities indicating priorities set in advance for the plurality of dictionaries, and stores one or more of the dictionary priorities obtained as a result of the search. And outputting the code based on the priority order corresponding to the code and the dictionary priority.