JPH0117192B2 - - Google Patents

Description

[Detailed Description of the Invention] The present invention provides a character recognition device using frequency information,
In particular, the frequency of use of characters can be used to distinguish between multiple categories, which is difficult to distinguish only based on similarity information or dissimilarity information, and the frequency of use information can be learned with high precision. The present invention relates to a character recognition device using frequency information, which can be used by replacing the memory according to each person or each theme if a replaceable memory is used.

Character recognition devices that have been put into practical use so far have been used to recognize alphanumeric characters, kana characters, and the like, as well as numerical values and product names. For this reason, the frequency of use of each input character was also approximately uniform. Therefore, efforts have been made to accurately recognize input characters without errors. This direction continued when it came to kanji recognition.

However, in the case of kanji recognition, the number of categories is enormous and there are an extremely large number of similar characters, making it extremely difficult to separate and identify them individually. However, in application fields where kanji are used as input characters, they are often used to recognize sentences and words. Consideration is being given to utilizing this fact to aid in discrimination based on surrounding characters and grammatical connections. That is, it is also considered to prepare a word dictionary in advance and use the word dictionary to help recognize unrecognizable characters from the characters before and after the characters become recognizable.

However, in the case of this method, it is necessary to prepare a word dictionary for each field of use, such as what kind of theme the input characters (or sentences) relate to, and the process of extracting words from the preceding and following characters is also required. It's not that easy.

The purpose of the present invention is to solve the above-mentioned problems, so to speak, by regenerating character usage frequency information for each person or each theme through learning, and assisting in the recognition of unrecognizable characters. The purpose is to improve the recognition rate. Therefore, using the frequency information of the present invention, the character recognition device uses the input character and the contents of the dictionary section to extract similarity information regarding the degree of similarity or mismatch to which category the input character corresponds. In a character recognition device equipped with a similarity extraction processing unit, a replaceable frequency table/memory circuit unit stores the frequency with which characters belonging to the category are input as input characters corresponding to each category; a frequency determination circuit unit that performs category determination based on frequency information read from the frequency table memory circuit unit for a plurality of categories that are similar to each other within a predetermined threshold;
and a frequency calculation circuit section that calculates the frequency of occurrence of the input character based on the result of recognizing the input character, and the contents of the frequency table memory circuit section can be updated based on the calculation results by the frequency calculation circuit section. The present invention is characterized in that an input character is recognized using at least both of the above-mentioned similarity information and the above-mentioned frequency information. This will be explained below with reference to the drawings.

The figure shows the configuration of an embodiment of the present invention. Reference numeral 1 in the figure is a similarity extraction processing unit, which uses standard feature information stored in a dictionary as well-known to extract similarity information (similarity or mismatch) regarding which category the input character corresponds to. ); 2 is a threshold determination circuit unit which determines whether a single category can be separated based on the similarity information extracted by the similarity extraction processing unit 1; 3-0 , 3-1, 3-2... are frequency table/memory circuit sections that store usage frequency information for each character that may be input, and can be replaced by each person or according to each theme. 4 is a frequency determination circuit unit which receives candidate category names for input characters that are determined to be unrecognizable and determines one of the candidate categories from the frequency information based on the contents of the frequency table/memory circuit unit 3-i. 5 is a frequency determination control circuit section which receives the candidate category name, indexes the frequency table/memory circuit section 3-i, and gives a determination instruction to the frequency determination circuit section 4; 6; 7 is a display memory circuit unit which stores the categories separated by the threshold value judgment circuit unit 2 or the category names separated by the frequency judgment circuit unit 4 and controls the display of the characters of the category; 8 is a display for displaying the characters of the above categories; 8 is a key section for modifying the characters displayed on the display 7 or for confirming the completion of the modification; 9
is a memory reading/frequency learning control section which instructs to read out the category name on the display memory circuit section 6 based on a correction end signal indicating that the characters displayed on the display 7 are correct; 10 indicates a frequency calculation circuit section which receives the category name on the display memory circuit section 6 and adds 1 to usage frequency information for the category.

Similarity extraction processing section 1
outputs what degree of similarity or degree of mismatch (hereinafter it is assumed that the degree of similarity is output) that the input character has with respect to which category. That is, for example, a predetermined number of categories are output in descending order of similarity, such as a% for category A, b% for category B, and so on. The threshold determination circuit unit 2 calculates α−β≧TH α−γ≧TH ( 1) Determine whether 〓 is true. If the condition is satisfied, the display memory circuit unit 6 is notified of the category name with the highest degree of similarity as separable.

If the condition of equation (1) is not satisfied, α≧TH′ β≧TH′ γ≧TH′ …(2) Notify 5. The frequency determination control circuit section 5 assigns the frequency table memory circuit section currently in use, for example 3-0, to the above category names A, B,
It accesses based on C and instructs the frequency determination circuit section 4 to start determination. At this time, from the frequency table memory circuit section 3-0, the category A
Frequency information a% about category B, frequency information b% about category C, frequency information c about category C
%... are sequentially notified to the frequency determination circuit section 4. The frequency determination circuit unit 4 determines the most frequently used category, for example, B, and notifies the display memory circuit unit 6 of the category name B.

The category name notified to the display memory circuit section 6 is stored in a memory (not shown), and the display memory circuit section 6 causes the display 7 to display characters corresponding to the category name. Whether or not this display is correct is checked by the operator by key input. If the answer is not correct, the category name on the display memory circuit section 6 is corrected by key input.

At the stage when the correct answer is obtained, memory read/
The frequency learning control unit 9 is activated. That is, the category name on the display memory circuit section 6 is read out and notified to the frequency calculation circuit section 10, and the content of the frequency table/memory circuit section 3-0 is added to the content of the frequency table/memory circuit section 3-0 for the category, and then the frequency table/memory circuit section is added again. Store in 3-0.

In this way, frequency information is used for characters that cannot be recognized only by the threshold value determination circuit section 2 (frequency information may also be used for characters that can be recognized) to improve the recognition rate. Make it. However, in general, the frequency of use of characters differs depending on each person or each theme. For this purpose, the contents of the frequency table/memory circuit section 3-i, whose accuracy has been increased by the learning effect as described above, can be changed by replacing the frequency table/memory circuit section 3-i for each person or theme. Make it possible and save it.
It is then set and used again when recognizing writings by the same person or writings on the same theme.

As described above, according to the present invention, even characters that cannot be correctly recognized only by the similarity extraction processing section can be recognized relatively easily. Further, since the accuracy of the contents of the frequency table memory circuit section 3-i is improved through learning, there is no need to prepare frequency information in advance. Furthermore, the frequency table memory circuit section 3-i uses a memory that can be replaced with, for example, a floppy disk, making it possible to enjoy the effects of the above-mentioned learning forever.

[Brief explanation of drawings]

The figure shows the configuration of an embodiment of the present invention. In the figure, 1 represents a similarity extraction processing section, 2 a threshold determination circuit section, 3 a frequency table/memory circuit section, 4 a frequency determination circuit section, 6 a display memory circuit section, and 10 a frequency calculation circuit section.

Claims (1)

[Claims] 1. In a character recognition device equipped with a similarity extraction processing section that extracts similarity information on the degree of similarity or mismatch to which category an input character corresponds using the input character and the contents of the dictionary section, a replaceable frequency table/memory circuit unit that stores the frequency with which characters belonging to the category are input as input characters, corresponding to a plurality of categories whose similar information is similar to each other within a predetermined threshold; In contrast, there is a frequency determination circuit section that performs category determination based on the frequency information read from the frequency table/memory circuit section, and a frequency calculation circuit section that calculates the frequency of occurrence of the input character based on the result of recognizing the input character. and configured to be able to update the contents of the frequency table/memory circuit section based on the calculation results by the frequency calculation circuit section, and recognize input characters by using at least both of the above similarity information and the frequency information. A character recognition device using frequency information, characterized in that: