JPH04151786A - Pattern recording device - Google Patents

Abstract

PURPOSE:To improve the recognition rate and to increase the recognition speed by storing feature patterns having a high frequency in use in dictionaries for first read and copying feature patterns of dictionaries for reread to dictionaries for first read at the time of extracting feature patterns corresponding to an input character or an input voice from dictionaries for reread. CONSTITUTION:When a feature pattern related to an input character type does not exist in dictionaries 16A to 18A for first read, the feature pattern of the input character type is collated with feature pattern groups in dictionaries 16B to 18B for reread. When the feature pattern related to the input character type exists in dictionaries 16B to 18B for reread, this feature pattern is copied to dictionaries 16A to 18A for first read; and feature patterns in dictionaries 16A to 18A for first read are successively extended as a set of specific pattern groups to reduce the recognition processing for dictionaries 16B to 18B for reread. Thus, the recognition object character type can be recognized by the recognition processing or the like of dictionaries 16A to 18A for first read, and dictionaries 16A to 18A for first read are optimized to improve the recognition rate and to increase the recognition speed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a pattern recognition device, and particularly to a pattern recognition device suitable for recognizing character type patterns such as characters and figures, or patterns related to voice information.

[Conventional technology]

For example, when recognizing printed or handwritten Chinese characters, the characteristics of the input characters are generally extracted and compared with a dictionary for recognition. In this case, a common general-purpose dictionary is used for unspecified authors or fonts, but if the similarity between authors or fonts is not very high, it is difficult to absorb differences between authors or fonts, and the recognition rate cannot be improved. However, the dictionary for recognition becomes enormous, which reduces recognition speed and recognition rate.

Therefore, conventionally, if we take character recognition as an example,
- As seen in Publication No. 155390, a means for discriminating typefaces by preparing a character dictionary for each typeface, or a special dictionary for each speaker for voice recognition and for each writer for handwritten character recognition. Measures have been taken. However, since there are a large number of authors and fonts, such a method would generally result in a huge dictionary, making it impractical.

In Japanese Patent Application Laid-Open No. 63-216390, when correcting a character whose recognition result is incorrect in an optical character recognition device, the image characteristics of the corrected character are reflected in a dictionary, and subsequent characters are about,
It is proposed that automatic correction be performed by re-recognizing the dictionary using a revised dictionary. However, with this method, there is a problem that the dictionary gradually becomes specialized to the author or typeface of the user of the device, and it becomes unable to deal with unspecified authors.

In Japanese Patent Application Laid-Open No. 1-173273, when recognition is performed for an unspecified number of people in a character or voice recognition method, templates corresponding to the writings of many authors are prepared in a dictionary, and a We have proposed a selective learning method for recognition dictionaries in which a template with a pattern most suitable for the individual is selected and used as a temporary dictionary by having the individual fill in samples of specified specific characters. However, this method is a proposal for a dictionary selection method, and does not essentially solve the problem of a dictionary construction method necessary for recognition.

Furthermore, in Japanese Patent Application Laid-Open No. 1-241689, a dictionary unit having a first dictionary storing frequently used character types and a second dictionary storing infrequently used character types is prepared, and reading information is stored in the dictionary section. A method of performing recognition processing based on the character type stored in a first dictionary, and re-recognizing the character to be read based on the character type stored in a second dictionary if the recognition result is incorrect recognition or unrecognized. is proposed. However, in this method, the character types included in the hierarchical dictionary are fixed, so it is not possible to respond to the frequency of appearance of character types that changes depending on the user's purpose of use.
Some users may place certain characters in a second dictionary that stores characters that are used less frequently even though they are used frequently, or vice versa. , there were problems such as the first dictionary storing frequently used character types.

Furthermore, Japanese Patent Application Laid-Open No. 1-267790 discloses a method of initializing the recognition dictionary of the personal character recognition section with the recognition dictionary of the general-purpose character recognition section in a character recognition method that includes a general-purpose character recognition section and a personal character recognition section. is proposed. However, with this method, the recognition dictionary of the personal character recognition unit is created as a copy of the recognition dictionary of the general-purpose character recognition unit, so some of the recognition dictionaries of the personal character recognition unit are frequently used by specific users. This results in a large number of unused character types, unnecessarily increasing the dictionary size, and contributing disadvantageously to recognition rate and recognition speed. Furthermore, since the results from both are collated during recognition execution, it is always necessary to run the general-purpose character recognition section and the personal character recognition section simultaneously, which requires twice the load on the device configuration.

[Problem to be solved by the invention]

As mentioned above, each of these methods has the problem that a realistic construction system for recognition dictionaries for the recognition of a large number of unspecified authors, characters in fonts, or voices of a large number of unspecified speakers has not been proposed. It was.

In general, a speech recognition dictionary for characters of an unspecified number of authors and fonts or for an unspecified number of speakers is created by collecting data of a large number of authors, fonts, or speakers and averaging these data. However, an averaged dictionary usually has a lower recognition rate than a dictionary of a particular author, typeface, or speaker. Therefore, there is a need for a system for configuring a recognition dictionary for characters of an unspecified number of authors, typefaces, or voices of an unspecified number of authors, which achieves the recognition rate of the dictionary section for a particular author, typeface, or speaker.

An object of the present invention is to provide a pattern recognition device that can quickly recognize patterns related to character types or voices of unspecified users.

[Means to solve the problem]

In order to achieve the above object, the present invention includes, as a first device, a dictionary for initial reading that stores characteristic patterns of a specific character type group unique to a user, and a characteristic pattern of a character type group common to each user. a rereading dictionary to be stored, an image information input means for inputting image information regarding character types, a pattern generation means for generating a pattern of character types from the image information inputted by the image information input means, and an input from the pattern generated by the pattern generation means. Feature extracting means for extracting features of character types, and pattern recognition control for comparing and matching the feature pattern of the input character type extracted by the feature extracting means with the feature pattern of a group of character types stored in at least one of the two dictionaries. and a display means for displaying an image of the recognition result of the pattern recognition control means, the pattern recognition control means collating the characteristic pattern of the input character type with the characteristic pattern group of the dictionary for initial reading, and displaying the recognition result in the dictionary. a first determination unit that determines whether or not a feature pattern corresponding to the input character type exists; and a feature pattern that corresponds to the input character type is extracted from the dictionary when a positive determination result is obtained by the first determination unit. a first image information output section that outputs image information regarding the character type of the feature pattern extracted by the first pattern extraction section; a second determining unit that compares the characteristic pattern of the character type with a group of characteristic patterns in the rereading dictionary to determine whether or not a characteristic pattern corresponding to the input character type exists in the dictionary; a second pattern extraction unit that extracts a feature pattern corresponding to the input character type from the dictionary when a determination result is obtained; and a second image that outputs image information regarding the character type of the feature pattern extracted by the second pattern extraction unit. an information output section; a pattern copying section that copies the characteristic pattern extracted by the second pattern extraction section into a dictionary for initial reading; and, when a negative judgment result is obtained by the second judgment section, image information regarding this judgment result; This is a pattern recognition device that includes a second image information output section that outputs a second image information output section.

As a second device, a dictionary for initial reading stores feature patterns of specific audio information specific to the user, a dictionary for rereading stores feature patterns of audio information common to each user, and audio information is inputted. A voice information input means, a pattern generation means for generating a pattern related to voice from voice information inputted by the voice information input means, a feature extraction means for extracting a voice feature from a pattern generated by the pattern generation means, and a feature extraction means. pattern recognition control means for comparing and collating the extracted voice characteristic pattern with a voice characteristic pattern stored in at least one of the two dictionaries, and a display means for displaying an image of the recognition result of the pattern recognition control means. The pattern recognition control means compares the characteristic pattern of the input voice with a group of characteristic patterns of the initial reading dictionary to determine whether a characteristic pattern corresponding to the input voice exists in the dictionary. a first determining unit; and a first determining unit that extracts a feature pattern corresponding to the input voice from the dictionary when a positive determination result is obtained by the first determining unit.
a first image information output section that outputs image information regarding the sound of the feature pattern extracted by the first pattern extraction section; and a first image information output section that outputs image information regarding the sound of the feature pattern extracted by the first pattern extraction section; a second determination unit that compares the pattern with a group of characteristic patterns in the rereading dictionary to determine whether or not a characteristic pattern corresponding to the input voice exists in the dictionary;
a second pattern extraction unit that extracts a feature pattern corresponding to the input audio from the dictionary when a positive determination result is obtained by the second determination unit; and an image related to the audio of the feature pattern extracted by the second pattern extraction unit; a second image information output section that outputs information; a pattern copy section that copies the characteristic pattern extracted by the second pattern extraction section into a dictionary for initial reading; and a second image information output section that outputs information;
The pattern storage device includes a third image information output section that outputs image information regarding the determination result when a negative determination result is obtained by the determination section.

A third device including the first or second device is provided with a plurality of dictionaries as first reading dictionaries, and includes first reading dictionary selection means for sequentially specifying the dictionaries as checking dictionaries in response to an operation. This is a configuration of a pattern recognition device.

A fourth device including the first or second device is provided with a plurality of dictionaries as rereading dictionaries, and further includes rereading dictionary selection means for sequentially specifying the dictionaries as checking dictionaries in accordance with an operation. This is a configuration of a pattern recognition device.

A fifth device including the first or second device is provided with a plurality of dictionaries as a number book for discontinuation, and includes first reading dictionary selection means for sequentially specifying the dictionaries as check dictionaries according to operations. The pattern recognition apparatus is provided with a plurality of dictionaries as rereading dictionaries, and a rereading dictionary selection means for sequentially specifying the dictionaries as checking dictionaries in response to an operation.

A sixth device including any one of the first to fifth devices
As the device, the first image information output section and the second image information output section constitute a pattern recognition device that outputs information specifying the dictionary from which the extracted feature pattern is stored, attached to the image information. It is something.

A seventh device including any one of the first to sixth devices
This apparatus is a pattern recognition apparatus including a characteristic pattern input means for storing a characteristic pattern corrected with respect to a recognition target pattern in a dictionary for initial reading.

[Effect]

When information regarding a character type to be recognized is input, image information regarding this input character type is converted into a pattern, and this pattern is first compared with a specific pattern stored in a dictionary for initial reading. When a feature pattern corresponding to the input character type exists in the first-reading dictionary, image information about that feature pattern is displayed, and when there is no feature pattern related to the input character type in the first-reading dictionary, the feature pattern of the input character type and the re-reading dictionary are displayed. A comparison with a group of feature patterns in the dictionary is performed. And when there is a feature pattern related to the input character type in the rereading dictionary.

Image information related to this input character type is displayed, and the feature patterns existing in the re-reading dictionary are copied to the first-reading dictionary, and the feature patterns in the first-reading dictionary are sequentially expanded as a set of specific patterns of specific things. , reducing recognition processing for the rereading dictionary.

This makes it possible to recognize the target character type through recognition processing for the first reading dictionary, optimizing the first reading dictionary, and improving the recognition rate and recognition speed. Since similar processing is performed on audio information, it is possible to improve the recognition rate and recognition speed for audio information as well.

Furthermore, by providing multiple dictionaries for initial reading or multiple dictionaries for rereading, it is possible to select recognition target patterns from a large amount of information and improve the recognition rate. can.

In addition, when the required pattern cannot be obtained in the recognition process, it is possible to optimize the first reading dictionary by performing a learning process that stores a specific pattern modified from the recognition target pattern in the first reading dictionary. Become.

〔Example〕 Hereinafter, one embodiment of the present invention will be described based on the drawings.

In FIG. 1, the pattern recognition device includes a keyboard 10,
Recognition control unit 11. Scanner 12, image buffer 13
, pre-processing unit 14, feature extraction unit 15, broad classification dictionary unit 16
, medium classification dictionary 17, small classification dictionary section 18, CRT 19
Each dictionary section 16, 17, and 18 includes a dictionary for major classification for cutting and continuing 16A, a dictionary for medium classification for cutting and continuing 17A, a dictionary for small classification for cutting and continuing 18A, and a dictionary for major classification for rereading. It is hierarchically divided into a dictionary 16B, a dictionary for medium classification for rereading 17B, and a dictionary for small classification for rereading 18B.

The keyboard 10 is configured as an input means for inputting various information and commands, and when an operation button on the keyboard 10 is operated, the command according to the operation button is transmitted to each section via the recognition control section 11. with CRT19
The content of the operation is displayed as an image on the screen. The scanner 12 follows instructions from the recognition control unit 11.
It is configured as an image information input means that optically scans characters on a form and outputs image information obtained by scanning, and the image information is stored in the image buffer 13. The image information stored in the image buffer 13 is subjected to preprocessing such as character extraction and normalization processing by a preprocessing section 14, so that a pattern of input characters is generated by operation of the scanner 12. That is, the image buffer 13 and the preprocessing section 14 are configured as pattern generation means for generating character patterns from image information. The pattern generated by the preprocessing section 14 is then transferred to the feature extraction section 15, where the features of the cut out characters are extracted. That is, the feature extracting section 15 is configured as a feature extracting means for extracting features of input characters.

Initial reading dictionary 16A, 17 among dictionary parts 16, 17, 18
A, 18A shows the characteristic patterns of specific types of characters such as characters and figures that are unique to the user.For example, characteristic patterns unique to each author or typeface are categorized into major, medium, and minor categories. The major classifications are stored in the first reading dictionary 16A, the middle classifications are stored in the first reading dictionary], 7A, and the small classifications are stored in the first reading dictionary 18A. On the other hand, rereading dictionary 16B.

17B and 18B contain feature patterns of character type groups common to each user, such as major classifications when character characteristic patterns having common characteristics among authors or fonts are divided into major classifications, medium classifications, and small classifications. are stored in the rereading dictionary 16B, those in the medium category are stored in the rereading dictionary 17B, and those in the small category are stored in the rereading dictionary 18B. That is, each of the rereading dictionaries 16B, 17B, and 18B stores data of a feature pattern group created by collecting data of a large number of authors or fonts and averaging these data. Further, the dictionaries 16A and 16B store feature patterns related to the external structure of characters, the dictionaries 17A and 17B store feature patterns related to the internal structure of characters, and the dictionary 18
Feature patterns related to the topological structure of characters are stored in A and 18B.

The recognition control unit 11, as a pattern recognition control means, uses the characteristic pattern of the input character type and the initial reading dictionaries 16A, 17A, 1
a first determining unit that determines whether or not a characteristic pattern corresponding to the input character type exists in the dictionary by comparing the characteristic pattern group of 8A; and when an affirmative determination result is obtained by the first determining unit; a first pattern extraction unit that extracts a feature pattern corresponding to the input character type from a dictionary; an image information output unit that outputs image information regarding the character type of the feature pattern extracted by the first pattern extraction unit; and a first determination unit. When a specified determination result is obtained, the feature pattern of the input character type is compared with the feature pattern groups in the rereading dictionaries 16B, 17B, and 18B to determine whether there is a feature pattern corresponding to the input character type in the dictionary. a second determination unit that determines whether the input character type is the same; a second pattern extraction unit that extracts a feature pattern corresponding to the input character type from the dictionary when the second determination unit obtains a positive determination result; a second image information output section that outputs image information regarding the character type of the feature pattern extracted by the second pattern extraction section; a pattern copy section that copies the feature pattern extracted by the second pattern extraction section to the initial reading dictionary 16A, 17A, 18A; The third image information output section is configured to output image information regarding the determination result when a negative determination result is obtained by the second determination section. Furthermore, each first reading dictionary 1 is used as a collation dictionary in response to commands from the keyboard 10.
Initial reading dictionary selection means for sequentially specifying 6A, 17A, and 18A, and rereading dictionaries 16B, 17B, and 1 as checking dictionaries.
A rereading dictionary selection means for sequentially specifying 8B is configured. Further, it constitutes a feature pattern input means for executing a learning process of storing feature patterns corrected for the recognition target pattern in the initial reading dictionaries 16A, 17A, and 18A.

In the above configuration, the keyboard 1 is
When a command is issued from 0, the character string on the form is optically scanned by the scanner 12, and image information regarding this character string is input into the image buffer 13. This image information is subjected to character extraction and size normalization in the preprocessing section 14, and a character pattern is generated. Then, the features of the cut out characters are extracted as feature codes by the feature extractor 15. Thereafter, through the process of the recognition determination unit 11, the feature extraction unit] 5 compares and matches the feature code related to the character feature pattern extracted by the feature code with the feature code related to the feature pattern stored in the initial reading dictionary 1.6A, 17A, 18A. will be held. When comparing and matching the two feature codes, the degree of difference between the two codes is defined as the "distance", and candidate characters are narrowed down according to the value of this distance. In other words,
Dictionaries for first reading 17A and 18A in order from first reading dictionary 16A
and executes processing to determine whether a feature code corresponding to the feature code of the input character exists in any of the dictionaries.

If there is a candidate character having a distance less than a certain threshold, the character closest to the input character is taken as the recognition result character, and a small number of candidate characters are output as discontinuation results in order of distance.

Furthermore, the recognition result characters and candidate characters are outputted along with image information for specifying from which initial reading dictionary the recognition result characters and candidate characters are output.

On the other hand, when there is no feature code corresponding to the feature code of the input character in the initial reading dictionaries 16A, 17A, 18A, the rereading dictionaries 16B', 17B.

A process of sequentially referencing the feature coats in 18B is executed.

In this case, as in the process described above, if there is a candidate character with a distance less than a certain threshold, the character closest to the input character is selected as the recognition result character, and then a small number of candidate characters are selected in order of distance. Output as the rereading result. Furthermore, the character feature codes obtained as a result of rereading are copied to the initial reading dictionaries 16A, 17A, and 18A, respectively. For example, when a candidate character is selected from the rereading dictionary 16B, the feature code of this candidate character is copied to the first reading dictionary 16A. Furthermore, identification information for specifying the dictionary from which the characters were selected is added to the image information of the recognition result characters and candidate characters and output.

Further, if there is no candidate character having a distance equal to or less than a certain threshold value even after the rereading process, the image information is outputted with a reject code attached thereto as unrecognizable. Image information regarding each rereading result is displayed as an image on the screen of the CRT 19.

When the discontinuation result or rereading result is displayed on the screen of the CRT 19, the identification information added to the identification result character and the candidate character are displayed, so it is easy to check whether the character displayed in the image is a character obtained by the discontinuation result. The operator can identify whether the characters are obtained from the rereading results. The operator then checks the results of the button recognition from the image displayed on the screen of the CRT 19, and if the recognition result is correct, leave it as it is; if the recognition result is incorrect, display the candidate character group and the image to be read. , if the correct character exists in the candidate category, select that character, and if the correct character does not exist in the candidate category or is a displacement, enter the correct character, for example as a character code, to correct misreading. It can be carried out.

At this time, a learning process is executed in which the feature pattern corrected for the recognition target pattern is stored in the initial reading dictionaries 16A, 17A, and 18A. This learning process is performed by the first reading dictionary 16A.
, 17A, and 18A are also executed as a new addition of a category, and if the newly added character code already exists, it is executed as an overlay on the category, for example, as a logical sum of old and new image information. As a result, the initial reading dictionaries 16A and 17A are used until the characters to be recognized are reread.

After rereading, the information on the characters that are not included in 18A is included in the dictionary, and when the next character appears on the form, rereading processing is not necessary. Therefore, even if the number of character types included in the initial dictionary is sufficiently small, by repeating this process appropriately, the initial reading dictionary 16A
, 17A, and 18A can store characteristic patterns of frequently used character types.

In this way, according to the present embodiment, the initial reading dictionaries 16A, 1
Frequently used feature patterns are stored sequentially from 7A and 18A, so first reading dictionaries 16A and 17A
, 18A improves the recognition rate, and it becomes possible to improve the recognition rate and recognition speed as an OCR system.

When the operator performs correction work, he assigns, for example, "Reread" to one of the 10 operation buttons on the keyboard, and when the operator performs the correction work, he assigns, for example, "Reread" to one of the 10 operation buttons on the keyboard. This is executed by assigning, for example, "reread" to an operation button, and inputting a signal for pressing this key button and information about the character to be corrected. Further, the learning process is performed only when the operator specifies the user dictionary creation mode out of the user dictionary creation mode and the user dictionary execution mode when the system is started. If execution of the learning process is selected by specifying such a mode, the operator can work while clearly grasping the execution state of the dictionary.

In addition, in the above embodiment, if a dictionary corresponding to the author or typeface is prepared for each category as a dictionary for discontinuation, the operator can select any dictionary according to the purpose of use.
○It becomes possible to further improve the recognition rate and recognition speed as a CR system. An example of selecting a dictionary for initial reading in this case is shown in FIG.

In addition, when the operator performs correction work, 1.
For example, it is possible to assign "Learning" to "Learning" and have the user select whether or not to perform learning by selecting this menu. As shown in the figure. In this case, the specification is performed using function keys assigned by the menu and a pointing device such as a mouse.

Furthermore, in the embodiment described above, a system that recognizes patterns of character types such as letters and figures has been described, but the embodiment can also be applied to systems that recognize speech information.

For example, an audio input device such as a microphone may be used instead of the scanner 12, and an audio speaker sofa may be used instead of the image buffer 13. Then, the audio signal inputted by the audio input device is converted into a TSP (T-ime Spectrum Pa), which is a pattern representing the temporal change in audio frequency.
ttern), and converts this into a binary pattern in which the vicinity of the TSP peak is set to (1) and the others are set to (0). Binary pattern (Binary Time)
Spectrum Pattern) is character recognition 2
It can be handled in the same way as a digitized image, and the preprocessing unit 14 cuts off the syllable body and normalizes the syllables, and this binarized speech information is processed by cutting out the syllable body and normalizing the syllables. It is extracted as a code by the feature extraction unit 15.

The subsequent processing is the same as the d recognition processing for character type patterns, so that speech patterns can be recognized.

〔Effect of the invention〕

As explained above, according to the present invention, frequently used feature patterns are stored in the first reading dictionary, and when feature patterns corresponding to input characters or input speech are extracted from the rereading dictionary, the rereading dictionary is stored. Since the feature patterns are copied to the first reading dictionary, the first reading dictionary is optimized and the processing required to recognize the rereading dictionary is reduced, thereby improving the recognition rate and recognition speed. becomes possible.

Furthermore, the rereading dictionary is stored in its original form, so there is no risk of deterioration due to learning, and even if the contents of the continuous dictionary are updated, the contents of the rereading dictionary can be reused.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram showing one embodiment of the present invention, FIG. 2 is a diagram showing a display example of a selection screen for a dictionary for initial reading, and FIG.
It is a diagram showing an example of a display in which learning J designation is realized by a menu on a screen on a CRT. DESCRIPTION OF SYMBOLS 10... Keyboard, 11... Recognition control part, 12.
...Scanner, 13...Image buffer, 14...
- Preprocessing unit, 15... Feature extraction unit, 16A... Dictionary for large classification for discontinuation, 16B... Dictionary for large classification for rereading, 17A... Dictionary for medium classification for discontinuation, 17B. ... Dictionary for intermediate classification for rereading, 18A... Dictionary for small classification for discontinuation, 18B... Dictionary for small classification for rereading, 19... CRT.

Claims (1)

[Claims] 1. An initial reading dictionary that stores feature patterns of specific character types among user-specific character types, a rereading dictionary that stores feature patterns of character types common to each user, and images related to character types. An image information input means for inputting information, a pattern generation means for generating a character type pattern from the image information input by the image information input means, and a feature extraction means for extracting characteristics of the input character type from the pattern generated by the pattern generation means. , a pattern recognition control means for comparing and matching the characteristic pattern of the input character type extracted by the characteristic extraction means with the characteristic pattern of a character type group stored in at least one of the two dictionaries; and a recognition result of the pattern recognition control means. and display means for displaying an image of the input character type, and the pattern recognition control means compares the characteristic pattern of the input character type with a group of characteristic patterns in the dictionary for initial reading to determine whether a characteristic pattern corresponding to the input character type exists in the dictionary. a first determination unit that determines whether or not the character type is inputted; a first pattern extraction unit that extracts a feature pattern corresponding to the input character type from the dictionary when the first determination unit obtains a positive determination result; a first image information output unit that outputs image information regarding the character type of the feature pattern extracted by the pattern extraction unit; and a first image information output unit that outputs image information regarding the character type of the feature pattern extracted by the pattern extraction unit; a second determination unit that determines whether or not a feature pattern corresponding to the input character type exists in the dictionary by comparing the pattern group; a second pattern extraction section that extracts a feature pattern corresponding to the input character type from among them; a second image information output section that outputs image information regarding the character type of the feature pattern extracted by the second pattern extraction section; and a second pattern extraction section. a pattern copying section that copies the feature pattern extracted by the first reading dictionary; and a second image information output section that outputs image information regarding the determination result when the second determination section obtains a negative determination result. A pattern recognition device consisting of. 2. An initial reading dictionary that stores characteristic patterns of specific voice information unique to the user, a rereading dictionary that stores characteristic patterns of voice information common to each user, and voice information input means that inputs voice information. a pattern generation means for generating a pattern related to speech from the speech information input by the speech information input means; a feature extraction means for extracting features of the speech from the pattern generated by the pattern generation means;
pattern recognition control means for comparing and matching the voice feature pattern extracted by the feature extraction means with the voice feature pattern stored in at least one of the two dictionaries; the pattern recognition control means compares the characteristic pattern of the input voice with a group of characteristic patterns of the initial reading dictionary to determine whether there is a characteristic pattern corresponding to the input voice in the dictionary. a first determination unit that determines whether the input voice is the same; a first pattern extraction unit that extracts a feature pattern corresponding to the input voice from the dictionary when the first determination unit obtains a positive determination result; a first image information output section that outputs image information regarding the sound of the feature pattern extracted by the section; and a feature pattern of the input sound and a group of feature patterns of the rereading dictionary when a negative judgment result is obtained by the first judgment section. a second determination unit that determines whether or not a feature pattern corresponding to the input voice exists in the dictionary by comparing the characteristic pattern with the input voice; a second pattern extraction unit that extracts a feature pattern corresponding to the input audio; a second image information output unit that outputs image information regarding the audio of the feature pattern extracted by the second pattern extraction unit;
a pattern copying section that copies the feature pattern extracted by the second pattern extraction section into a dictionary for initial reading; and a third pattern copying section that outputs image information regarding the judgment result when a negative judgment result is obtained by the second judgment section. A pattern storage device comprising an image information output section. 3. Equipped with multiple dictionaries for first reading,
3. The pattern recognition device according to claim 1, further comprising a first reading dictionary selection means for sequentially specifying the dictionary as a reference dictionary in response to an operation. 4. Equipped with multiple dictionaries for rereading,
3. The pattern recognition device according to claim 1, further comprising rereading dictionary selection means for sequentially specifying the dictionary as a reference dictionary in accordance with an operation. 5. Equipped with multiple dictionaries for first reading,
A first-reading dictionary selection means for sequentially specifying the dictionary as a collation dictionary according to an operation, a plurality of dictionaries as a rereading dictionary, and a rereading dictionary for sequentially specifying the dictionary as a collation dictionary according to an operation. 3. The pattern recognition device according to claim 1, further comprising dictionary selection means. 6. Claims 1, 2 and 3, wherein the first image information output section and the second image information output section output image information along with information specifying a dictionary from which the extracted feature pattern is stored. 6. The pattern recognition device according to 4 or 5. 7. The pattern recognition apparatus according to claim 1, further comprising a feature pattern input means for storing a feature pattern modified with respect to the recognition target pattern in a dictionary for initial reading.