JP4528705B2 - Character recognition device and character recognition method - Google Patents

Description

  The present invention relates to a character recognition device used for recognizing a program telop in a digital video recorder, for example.

  In the post-processing step of character recognition, knowledge processing such as error correction and phrase replacement using a knowledge dictionary is performed. For example, if it is an apparatus which recognizes an address, the word collation process with the knowledge dictionary which registered the address of the whole country is performed. Further, if the device reads a document, word matching processing is performed using a general word dictionary as a knowledge dictionary.

  However, since these dictionaries are usually prepared in advance, there is a problem that it is difficult to create dictionaries when necessary words cannot be assumed.

  In character recognition, it is better to correct words that are likely to be misread, and it is better not to correct words that are less likely to be misread.

  As a response to this, in conventional word processing, it is necessary to delete words that are likely to be erroneously corrected in advance from the knowledge dictionary. Therefore, if the recognition unit is changed, it is necessary to review the knowledge dictionary was there.

  For example, when a digital video recorder or the like is equipped with a telop recognition function using word matching processing, which is one of knowledge processing, a dictionary required as a knowledge dictionary for performing word matching depends on the user to use. Change.

This is because the tendency of the words appearing changes according to the program recorded by the user. In such a case, it is necessary to improve the performance by correcting the knowledge dictionary.
As a simple correction method for correcting the knowledge dictionary, there is a method of adding a newly appearing word, but this does not necessarily improve the performance of the word matching process.

  This is because, for example, when a word with high similarity such as “tee” and “fee” is in the dictionary, it is not always possible to mechanically determine which one is appropriate. For example, a short word such as “car” This is because many words in the knowledge dictionary are matched, so that when a misreading occurs, it is easy to determine that it is a correct answer of another word.

  For this reason, the knowledge dictionary is generally classified according to the recognition target and the appearance frequency is considered, but these are set in advance, and the tendency of how to use characters for each user is always considered. It is not a thing.

  The present invention has been made to solve such a problem, and provides a character recognition device and a character recognition method capable of improving knowledge performance according to a certain tendency and thereby improving character recognition accuracy. It is an object.

In order to solve the above-described problem, the character recognition device of the present invention includes a telop recognition unit that recognizes characters of a television program, a plurality of words, and a television program genre as a condition for classifying each word. A knowledge dictionary storing information and a statistical value as a basis of evaluation including the number of occurrences of the word in each genre, and electronic program guide information of the television program is acquired from an external information providing means, and the electronic program from the guide information, the preset information including the genre name of the television program, and extracted as a keyword for extracting a word to be added to the dictionary for matching, extracting words from the knowledge dictionary by the keyword extracted The additional information given to the word is examined, the word obtained from the electronic program guide information in the examined additional information, the number of appearances of this word and the genre A collation dictionary creating means for creating a dictionary for collation in which the appearance rate calculated from the number of matches is matched; a word recognized by the telop recognition unit; and the dictionary for collation based on the keyword The word processing result of the word recognition result is calculated by comparing the words extracted from the words and calculating the similarity between them, and the word evaluation value calculated by the matching processing means is preset. And a knowledge dictionary updating means for updating the word used for the telop recognition of the knowledge dictionary and its additional information and statistics when the threshold is exceeded.

The character recognition method of the present invention is a telop recognition unit, based on an evaluation including a plurality of words, additional information including a genre of a television program as a condition for classifying each word, and the number of appearances of the word in each genre. In a character recognition method by a character recognition device comprising a knowledge dictionary storing statistical values, a collation dictionary creating means, a collation processing means, and a knowledge dictionary updating means, the telop recognition unit recognizes characters of the telop of the television program When the word the collation dictionary creation means, adding the electronic program guide information of a television program obtained from the outside information providing means, the preset information including the genre name of the television program, the dictionary for matching Is extracted as a keyword for extracting, and a word is extracted from the knowledge dictionary by the keyword, and is assigned to the extracted word The additional information is examined, and the collation dictionary is created by associating the word obtained from the electronic program guide information with the examined additional information and the appearance rate calculated from the number of appearances of this word and the number of genre comparisons. Comparing the word recognized by the telop recognition unit with the word extracted from the dictionary for matching based on the keyword and calculating the similarity between them, Calculating an evaluation scale for the word, and when the evaluation scale of the word exceeds a preset threshold, the knowledge dictionary updating means updates the word used for telop recognition of the knowledge dictionary, its additional information, and a statistical value. Steps.

In the present invention, keywords for extracting words to be added to a matching dictionary and words to be added to a matching dictionary are acquired from electronic program guide information of a television program acquired from an external information providing means, and knowledge is acquired by using the keywords. Create a collation dictionary based on the words extracted from the dictionary and the words to be added to the collation dictionary , collate the words obtained by telop character recognition with the collation dictionary information, and make similarities to each other Is used to calculate the evaluation scale for the word of the character recognition result, and when the evaluation scale of the word exceeds a preset threshold, the corresponding information in the knowledge dictionary is updated. That is, the knowledge about the recognition target is acquired from the outside, the performance of the knowledge processing is improved using the acquired knowledge, and the knowledge dictionary is updated from the result of the knowledge processing. Thereby, according to a certain tendency, the performance of knowledge processing can be improved and the character recognition accuracy can be improved.

  As described above, according to the present invention, it is possible to improve knowledge performance in accordance with a certain tendency, thereby improving character recognition accuracy.

(First embodiment)
Embodiments of the present invention will be described below in detail with reference to the drawings.
FIG. 1 is a diagram showing the configuration of a digital video recorder according to the first embodiment of the present invention.

  The digital video recorder 1 of the first embodiment includes a TV tuner unit, a digital video recording unit, a display operation unit, etc., which are basic functions as a recorder, as well as a knowledge dictionary 11 and a collation dictionary creation as shown in FIG. Part 12, collation dictionary 13, telop recognition part 14, word collation processing part 15, knowledge dictionary update part 16, update information 17, television monitor 18 as an output part of processing results, external information acquisition part 19, etc. Knowledge dictionary update means is provided.

  The TV tuner unit receives an analog broadcast or a digital broadcast television program. The TV tuner unit outputs the received video information of the TV program to the telop recognition unit 14. The digital video recording unit digitally records the television program received by the TV tuner unit on any storage medium such as a hard disk device, a DVD, a DV tape, or a flash memory. The display operation unit accepts an input operation from the user. Input operations include channel switching operations, volume adjustment operations, program recording reservation operations, program recording operations, digital video playback operations, and the like.

  The digital video recorder 1 includes hardware such as a CPU, a memory, and a hard disk device. The collation dictionary 13 and the update information 17 are information temporarily created and stored in a memory or a hard disk device. The knowledge dictionary 11 is stored in a hard disk device or the like.

  The external information acquisition unit 19, knowledge dictionary 11, collation dictionary creation unit 12, collation dictionary 13 as a temporary word collation dictionary, word collation processing unit 15, knowledge dictionary update unit 16, etc. Since the dictionary 11 is updated, it is called a post-processing unit 10. The post-processing unit 10 can be realized by hardware or software. The telop recognition unit 14 and the television monitor 18 can be said to be external components with respect to the post-processing unit 10. That is, the telop recognition unit 14 is a recognition unit existing outside.

  The external information acquisition unit 19 accesses, for example, an EPG server on the Internet, and downloads and acquires electronic program information A (hereinafter referred to as EPG information A) from the EPG server. The EPG information A is external information published on the Internet, and includes information related to a word or a word genre. Specifically, the external information acquisition unit 19 corresponds to a device that accesses the Internet, such as a LAN card, router, or wireless LAN connected to an optical modem.

  The collation dictionary creating unit 12 creates collation dictionary creating means for creating the collation dictionary 13 used in the word collation processing unit 15 using the EPG information A acquired from the outside by the external information acquiring unit 19 and the information of the knowledge dictionary 11. Function as.

  The telop recognition unit 14 recognizes the telop from the video information of the news program input from the TV tuner unit, for example, while the user is watching the news program, etc., and the name of the moderator of the news program is written as the recognition result. Get as a column. In music programs, singer names, lyrics, etc. are obtained as telop recognition results.

  As the telop recognition technology, an existing character recognition technology for recognizing a character string (text data) from an image (image data) is used. Specifically, the telop recognition unit 14 detects and cuts out a range of characters from the image, performs matching with the character image cut out using a dictionary in which the character image and text data are associated, and the character images match or Similar text data is read from the dictionary and output.

  The knowledge dictionary update unit 16 performs knowledge processing, that is, word matching processing, on the recognition result of the telop recognized by the telop recognition unit 14 using the matching dictionary 13 created by the matching dictionary creation unit 12. Then, update information 17 for updating the knowledge dictionary 11 is evaluated by evaluating the character recognition result by the telop recognition unit 14, and the evaluation result is output to the television monitor 18, for example. Further, the word collation processing unit 15 temporarily stores the update information 17 created by the word collation processing (knowledge processing) in the memory and outputs it by reading from the knowledge dictionary updating unit 16.

  Hereinafter, in this digital video recorder 1, description will be given by taking as an example a case where the telop recognition unit 14 as means for performing knowledge processing and EPG information A published by an EPG server on the Internet as external information are downloaded and used. .

  The knowledge dictionary 11 is a dictionary database including words already registered among words to which genre information used for knowledge processing is added.

  As shown in FIG. 2, in the knowledge dictionary 11, for each word, the genre in which the word appears, the number of appearances of the word in each genre, the number of substitutions for the word, and the substitution for the word are incorrect. Is stored as a base value for evaluation such as the number of times (hereinafter referred to as the number of erroneous substitutions).

Here, the number of replacements with the word is the number of times the replacement was performed because the portion determined to be the word in the recognition result is different from the word, and the number of replacements is the telop of the replacements. This is the number of times that the recognition unit 14 has made an erroneous replacement.
As described above, the knowledge dictionary 11 has information for each genre of words. However, the knowledge dictionary 11 is not necessarily provided for each genre, and a plurality of genres may be handled collectively. Further, the knowledge dictionary 11 holds the number of times that each genre is determined to be the genre and collated.

  The knowledge dictionary 11 according to the present embodiment includes the above-described contents. However, the contents included in the knowledge dictionary 11 are not limited to the above, and may include, for example, the last appearance time of each word. A histogram that statistically analyzes the number of appearances every month or the like may be included. Further, the condition for classifying (or classifying) the words in the knowledge dictionary 11 is not limited to the genre but may be one of the conditions for classifying the words by adding name information such as a person who easily utters the words. Good.

  The collation dictionary creation unit 12 adds from the knowledge dictionary 11 to the collation dictionary 13 from the EPG information A downloaded by the external information acquisition unit 19 from the EPG server on the Internet using the character string or word of the recognition result as a key. A keyword for extracting a word and a word to be added to the collation dictionary 13 are obtained, and a collation dictionary 13 is created based on the word extracted from the knowledge dictionary 11 by the keyword, the genre, and the word to be added to the collation dictionary 13. The word collation processing unit 15 has a function of providing the collation dictionary 13.

  In this embodiment, the keyword is the genre of the program (news, sports, economy, society, etc.), and the word added to the matching dictionary 13 is the name of the performer of the program, but the keyword itself may be used as the keyword. In addition to this, for example, a program name or broadcast date and time may be used. Further, as a word to be added to the collation dictionary 13, not only the name of a program performer but also other information such as a word extracted by syntactic analysis from a program content description sentence may be used.

  That is, the collation dictionary creation unit 12 uses a word or genre obtained by character recognition as a key, and obtains a character string or word and these character string or word from the knowledge dictionary 11 and an EPG server which is an external information providing means. It functions as a collation dictionary creating means for acquiring information to be classified and creating a collation dictionary.

  Here, with reference to FIG. 3, the creation operation of the collation dictionary 13 in the digital video recorder 1 will be described. 3 uses the genre “sports” as a keyword, the name of the program performer “Tokyo Taro” as a word to be added to the collation dictionary 13, and the knowledge dictionary 11 shown in FIG. It is a figure for demonstrating the flow which produces.

  As shown in FIG. 3, the collation dictionary creation unit 12 first examines the appearance genre information added to each word in the knowledge dictionary 11 to extract words including “sports”, and the number of collations for the genre “sports”. A collation dictionary 13 is created using the extracted word information.

  The appearance rate is the number of occurrences of words / the number of genre matching, the replacement rate is the number of replacements of words / the number of appearances of words, and the error replacement rate is the ratio of the number of word replacements / the number of word replacements (% ) By converting.

  Next, an appropriate appearance rate / replacement rate / error replacement rate is given to the word specified as a word to be added to the collation dictionary 13 and added to the original collation dictionary 13. Thus, a new collation dictionary 13 is created.

  Note that the collation dictionary 13 is not limited to the above format, and other information such as the last appearance time may be used. For example, as a method of calculating each ratio, other denominators may be used as the number of appearances of the genre. A method may be used. In this embodiment, the calculation is performed as described above, but it is also possible to simply extract a word including “sports” in the appearance genre information.

  The word collation processing unit 15 uses the new collation dictionary 13 created by the collation dictionary creation unit 12 for the character string “Tokyo Dairo Dai!” In the telop recognition result of the telop recognition unit 14. By performing knowledge processing, that is, word collation processing, the telop recognition result is evaluated, and update information 17 for updating the knowledge dictionary 11 is created.

  In this embodiment, an example of word matching processing based on dynamic programming is described as knowledge processing. However, other processing such as morpheme analysis is performed to divide into morpheme units, and word matching processing is performed in morpheme units. You may go.

  Dynamic programming is a method generally used when calculating the similarity of two series data. The most basic calculation method of dynamic programming is shown in FIG.

  As shown in FIG. 4, cost_u (n) is a cost that the nth character of the collation destination is a character that is not assigned to any character of the collation source, and cost_r (n) is the nth character of the collation source The cost that the character is a character that cannot be assigned to any character of the collation destination, and cost_p (n, m) is the cost that the nth character of the collation source is assigned to the mth character of the collation destination.

  The dynamic programming sequentially performs the above calculation from the lower left point to the upper right point, calculates the minimum cost required to reach each point, and the path that generates the minimum cost. This is a method of using a route as a matching result.

  Therefore, if the minimum cost at the uppermost rightmost point is calculated, it is possible to obtain the portion closest to the collation destination among the collation sources. This process is hereinafter referred to as matching.

  In the present embodiment, each word registered in the collation dictionary 13 is a collation destination in FIG. 4, and the telop recognition result recognized by the telop recognition unit 14 is the collation source in FIG.

のように表すことができる。
但し、照合元（認識結果）における各文字の類似度および大きさは、テロップ認識部１４から与えられるものとする。また、Ｃｏｓｔ（１，１）＝０とする。 In addition, each cost function for obtaining a cost is represented by assuming that the nth element of the collation source (recognition result) is Y (n) and the m-th character of the collation destination (word of the collation dictionary 13) is D (m), for example.

It can be expressed as
However, the similarity and size of each character in the collation source (recognition result) are given from the telop recognition unit 14. Further, it is assumed that Cost (1, 1) = 0.

という式で表すことができる。但し、各文字の出現率は、番組中に該文字が出現する比率、正読率は、該文字であると認識した結果が正しい確率、消滅率は、認識対象の中に存在する該文字が文字認識の結果に含まれない確率である。 However, the calculation method of each cost function is not limited to the above, and may be calculated using, for example, a database that holds information on each character as shown in FIG.
In this case, if S (n) is the nth character of the collation source (recognition result) and D (m) is the mth character of the collation destination (word of the collation dictionary 13), for example

It can be expressed by the formula However, the appearance rate of each character is the rate at which the character appears in the program, the correct reading rate is the probability that the result of recognition as being the character is correct, and the disappearance rate is the rate at which the character existing in the recognition target is Probability not included in the result of character recognition.

  In FIG. 6, the word collation processing unit 15 uses the cost function of the above formula (2) to change the word w “Tokyo Taro” in the collation dictionary 13 to the telop recognition result “Tokyo Dairou Dai!”. It is a figure which shows the result of matching.

  In FIG. 6, the bold arrow indicates the final matching result, and the entry “Tokyo Taro” in the matching dictionary 13 matches the first character to the fourth character of the recognition result “Tokyo Dairo Dai!”. The matching cost is 3.0.

のように定義できる。ここでαは定数である。また、コストＦ（ｗ）の計算には、例えば該単語の最終出現時刻などの情報を用いてもよい。 FIG. 7 is a diagram showing in detail a part of the calculation process for obtaining the matching result shown in FIG. The cost obtained as described above is set as the calculation cost C (w) of each word. The final cost F (w) of each word is calculated using the appearance rate, replacement rate, error replacement rate, and calculation cost C (w) of the word included in the new matching dictionary 13. Cost F (w) is, for example,

Can be defined as Here, α is a constant. For example, information such as the last appearance time of the word may be used for calculating the cost F (w).

  In this case, there is a method such that a new F (w) is obtained by multiplying the above expression (3) by the time from the last appearance time to the present time.

のような規則によって選択する。単語Ｘをマッチング結果と称す。
単語照合処理部１５は、単語Ｘの認識結果「東京大郎大活躍！」に対するマッチング位置を調べ、該位置の認識結果が単語Ｘと異なれば、該当部分を単語Ｘに置換するなどの知識処理を行い、該知識処理の結果を出力する。 Next, the word collation processing unit 15 extracts the word X from all the words w registered in the collation dictionary 13, for example.

Select according to the rules. The word X is referred to as a matching result.
The word collation processing unit 15 checks the matching position for the recognition result of the word X “Tokyo Dairo Dai!”, And if the recognition result at the position is different from the word X, the knowledge processing such as replacing the corresponding part with the word X. And outputs the result of the knowledge processing.

  In the above example, since the word X is “Tokyo Taro” and the content of the matching position in the recognition result is “Tokyo Taro”, the portion “Tokyo Taro” in the recognition result is changed to the word X “Taro Taro”. The entire character string replaced is “Tokyo Taro Great!”.

  Finally, the word collation processing unit 15 creates update information 17 for the knowledge dictionary 11 as shown in FIG. 8, for example, from the matching result and the replacement information that has been replaced, and stores it in a memory or knowledge dictionary update unit 16 is output. As shown in FIG. 8, the update information 17 in this embodiment includes the word “Sports” in which the word “Tokyo Taro” has appeared, “Yes”, whether replacement has been performed, “Yes”, and whether the replacement has been correct. “Yes” or the like is set.

When the information for each character as shown in FIG. 5 is used, the knowledge dictionary update unit 16 updates the information for each character based on the matching result by the word matching processing unit 15.
When correcting the character, the knowledge dictionary updating unit 16 corrects the character by using, for example, the number of appearances / corrected readings / erasures as information for each character and the total number of executions of collation as other information.

By using the above information, the appearance rate can be defined as the number of appearances / the total number of collations, the correct reading rate can be defined as the number of correct readings / the number of appearances, and the disappearance rate can be defined as a percentage conversion of the number of disappearances / the number of appearances.
In addition, the information in the knowledge dictionary 11 is updated with the number of appearances for all characters included in the replaced word, and correct reading for all characters that are not corrected by the word matching processing unit 15 among the above. The number of erasures is added to the character added to the replacement source word by the word matching processing unit 15 among the characters included in the replaced word, and the total number of matchings is unconditionally added by 1, respectively. Can be realized.

  In FIG. 8, the knowledge dictionary update unit 16 first adds / updates a new word to the knowledge dictionary 11. In this case, the word newly added to the knowledge dictionary 11 is “Tokyo Taro”, and the genre is “sports”.

  First, the number of verifications of the genre information included in the update information 17 is increased by 1. As a result, the number of collations initially set to “100” in the knowledge dictionary 11 is incremented by 1 and updated to “101”.

  Next, the knowledge dictionary update unit 16 searches the knowledge dictionary 11 for a word, and when the word exists in the knowledge dictionary 11, increases the number of appearances of the word by one. If the replacement has not been performed, the update process ends here. That is, the knowledge dictionary update unit 16 sets the number of appearances to “1” according to the setting of “Yes” or “Yes” in the update information 17 for the word “Tokyo Taro” and the genre “sports” newly added to the knowledge dictionary 11. "Is set.

  Subsequently, the knowledge dictionary update unit 16 increases the number of replacements of the word by 1 when the replacement is performed. That is, the knowledge dictionary update unit 16 sets the number of replacements to “1” according to the setting of “Yes” or “Yes” in the update information 17.

  Furthermore, the knowledge dictionary updating unit 16 increases the number of erroneous replacements by 1 when the replacement is incorrect. That is, the knowledge dictionary updating unit 16 keeps the number of erroneous replacements at “0” because there is no erroneous replacement according to the setting of whether the replacement is correct or “Yes” in the update information 17. The knowledge dictionary 11 can be updated as described above.

If the word exists in the knowledge dictionary 11, it is not a new word, so the number of appearances is set to “0”, and the number of replacements and the number of erroneous replacements are set with reference to the update information 17. . This completes the word addition / update.
The update of the knowledge dictionary 11 is not limited to the above. For example, when the last appearance time of each word is used, the last appearance time of the word may be set to the time when the update process is performed.

Moreover, the knowledge dictionary update part 16 deletes a word. Below, an example of the addition process in the case of deleting a word is shown.
The knowledge dictionary update unit 16 first checks whether the number of words included in the knowledge dictionary 11 exceeds the maximum value of words in the knowledge dictionary 11.

のように定義できる。 If the number of words does not exceed the maximum value of words in the knowledge dictionary 11, the knowledge dictionary update unit 16 ends the deletion of words. Further, the knowledge dictionary update unit 16 calculates an evaluation scale T (w) for each word w when the number of words exceeds the maximum value of the words in the knowledge dictionary 11. Assuming that the evaluation scale T (w) is a set of genres given to the word,

Can be defined as

のような規則によって選択し、該単語Ｚを辞書から削除する。 Next, the word Z to be deleted from all the words w is, for example,

And the word Z is deleted from the dictionary.

The above processing is repeated until the number of words included in the knowledge dictionary 11 is less than the maximum value.
In this embodiment, deletion is performed by using the restriction on the number of words. However, deletion is performed based on other criteria such as deletion when the value of the evaluation scale T (w) is below a certain value. Also good.

  As described above, according to the digital video recorder 1 of the present embodiment, according to the preference of the program for each viewer regarding the information for each genre of the TV program when the viewer who is the user views the TV program. By updating the knowledge dictionary 11, knowledge processing more adapted to the genre becomes possible, and as a result, the performance of knowledge processing can be improved.

  Further, by changing the knowledge dictionary 11 according to the contents of the program selected by the viewer and the word matching result of the word matching processing unit 15, the viewer's preference can be taken in.

That is, since the knowledge dictionary 11 is updated according to the viewer's selection and the word matching result (knowledge processing result), the word recognition accuracy is improved according to the contents selected by the viewer and the matching characteristics of the word matching processing unit 15. can do.
As a result, the knowledge performance can be enhanced in accordance with a certain tendency (in this example, the preference of the program viewed by the viewer), and thereby the character recognition accuracy can be improved.

In addition, this invention is not limited only to the said embodiment.
In the above embodiment, the case where the telop recognition of a program is performed in the digital video recorder 1 has been described. However, the present invention can also be applied to, for example, a face recognition by a camera.

It is a figure which shows the structure of the digital video recorder of one Embodiment of this invention. It is a figure which shows an example of information, such as a word and a genre, hold | maintained at a knowledge dictionary. It is a figure for demonstrating the process of the dictionary creation part for collation. It is a figure which shows an example of the calculation method in a dynamic programming. It is a figure which shows an example of information when a word collation process part collates the word. It is a figure which shows the word collation result by a dynamic programming. It is a figure which shows a part of calculation process by a dynamic programming. It is a figure which shows the update result of a knowledge dictionary.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Digital video recorder, 10 ... Post-processing part, 11 ... Knowledge dictionary, 12 ... Collation dictionary preparation part, 13 ... Collation dictionary, 13 ... Collation dictionary, 14 ... Telop recognition part, 15 ... Word collation process part, 16 ... Knowledge dictionary update unit, 17 ... Update information, 18 ... TV monitor, 19 ... External information acquisition unit.

Claims (6)

A telop recognition unit that recognizes characters in TV program telops;
A knowledge dictionary storing a plurality of words, additional information including a genre of a television program as a condition for classifying each word, and statistical values serving as a basis for evaluation including the number of appearances of the word in each genre;
Acquires electronic program guide information of the television program from an external information providing device, from the electronic program guide information, the preset information including the genre name of the television program, and extracts the words added to the dictionary for matching And extracting the word from the knowledge dictionary by the keyword, examining the additional information attached to the extracted word, the word obtained from the electronic program guide information to the additional information examined, A matching dictionary creating means for creating the matching dictionary in which the number of appearances of the word and the appearance rate calculated from the number of matching of the genre are associated;
By comparing the word recognized by the telop recognition unit with the word extracted from the matching dictionary based on the keyword and calculating the similarity between them, an evaluation scale for the word of the character recognition result is obtained. Collation processing means to calculate;
A knowledge dictionary updating means for updating the word used for telop recognition of the knowledge dictionary and its additional information and statistical values when the word evaluation scale calculated by the matching processing means exceeds a preset threshold value. A character recognition device. The knowledge dictionary update means
2. The character recognition apparatus according to claim 1, wherein the knowledge dictionary is updated by adding the word obtained from character recognition and / or electronic program guide information and the additional information to the knowledge dictionary. The knowledge dictionary updating means
The character recognition device according to claim 2, wherein a statistical value of the word is increased when the knowledge dictionary is updated. Knowledge in which a telop recognition unit stores a plurality of words, additional information including a genre of a television program as a condition for classifying each word, and statistical values serving as a basis for evaluation including the number of appearances of the word in each genre In a character recognition method by a character recognition device comprising a dictionary, a collation dictionary creation means, a collation processing means and a knowledge dictionary update means,
Recognizing the telop of the television program by the telop recognition unit;
Extracting the collation dictionary creating means, from the electronic program guide information of a television program obtained from the outside information providing means, the preset information including the genre name of the television program, the word is added to the dictionary for matching And extracting the word from the knowledge dictionary by the keyword, examining the additional information given to the extracted word, and adding the word obtained from the electronic program guide information to the examined additional information Creating a dictionary for matching, in which the number of appearances of the word and the appearance rate calculated from the number of matching of the genre are associated with each other;
By comparing the word recognized by the telop recognition unit with the word extracted from the matching dictionary based on the keyword and calculating the similarity between them, an evaluation scale for the word of the character recognition result is obtained. A calculating step;
The knowledge dictionary updating means updates the word used for telop recognition of the knowledge dictionary and its additional information and statistical values when the word evaluation scale exceeds a preset threshold. Character recognition method. The knowledge dictionary updating means
5. The character recognition method according to claim 4, wherein the knowledge dictionary is updated by adding the word obtained from character recognition and / or electronic program guide information and the additional information to the knowledge dictionary. The knowledge dictionary updating means
6. The character recognition method according to claim 5, wherein a statistical value of the word is increased when the knowledge dictionary is updated.

Images